Home

Basics of Image Processing and Analysis

1. 140 163 App 1 75 List of accompanying PDF 141 1 6 4 App 2 Measurement Options 142 1 6 5 App 3 Edge Detection Principle 147 1 6 6 App 4 Particle Tracker manual 149 1 6 7 App Particle Analysis 42x xx mue oS 163 1 6 8 App 6 Image Processing and Analysis software script ing ANC MACE ud ode Bey da ne dU Rode eds 165 1 669 App 7 Macro for Generating Striped images 167 1 6 10 App 8 Deconvolution Exercise aaa 168 EMBL CMCI Image Basic Course 1 1 Basics of Basics 1 1 Basics of Basics Handling of digital images in scientific research requires knowledge on the characteristics of digital images A digital image translates to numbers and is hence a quantitative signal by nature In this section we learn the very basics of the numerical nature of digital images Inappropriate handling not only lowers the quality of your analysis but it could also be possible that your processing is considered as a manipulation of data For this lat ter point please also refer to Rossner and Yamada 2004 There are some limits on acceptable image processing to maintain the scientific validity Standards on scientific image processing could be found in Digital Imag ing Ethics by Cromey 2007 11 1 Digital image is a matrix of numbers A digital image we display on a computer screen is made up of pixels We can see individual pixel by zooming up the image using the magnifying
2. 114 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series 1 5 Analysis of Time Series In this section we learn how to analyze dynamics using image sequences A time series of digital images usually called stack contains temporal dynamics of position and intensity and kinetics can be obtained from these data In general there are three types of dynamics 1 Position does not change but intensity changes over time 2 Position changes but the intensity does not change 3 Both Position and Intensity change over time An example of type 1 is the measurement of cargo transport dynamics in vesicle trafficking Hirschberg et al 1998 Transition of protein local ization from ER to Golgi then to the plasma membrane was measured over time by measuring the signal intensity in each statically positioned compartment This type of technique has evolved to various sophisticated methods based on the same principle measurement of signal intensity at a constant position Type 2 corresponds to the measurement of movement or object tracking and an example is the single particle tracking of mem brane surface proteins Murase et al 2004 An example of type 3 mea surement is the measurement of chemotaxis related protein accumulation dynamics during the Dictyostelium cell migration Dormann et al 2002 Analysis of type 3 dynamics is more specific and advanced so refer to other literature Miura 2005 15 1 Differe
3. You can now go ahead with the linking by clicking OK The progress of the algorithm will be displayed in the main Image Status Bar ImageJ File Edit Image Process Analyze Plugins Window Help Eoee ane Pm eo Detecting Particles in Frame 15 144 Third Step viewing the results After completing the particle tracking the result window will be displayed Click the Visualize all Trajectories button to view all the found trajectories File Edit Image Process Analyze Plugins Window Help E oeoa els ela K Results View Preferences Relink Particles Configuration Kernel radius 3 Cutoff radius 0 0 Percentile 0 6 4 Particle Tracker DONE Found 110 Trajectories All Trajectories Visualize All Trajectories Focus on Selected Trajectory Focus on Area Save Full Report Selected Trajectory Info Trajectories in Area Info This window displays an overview of all 110 found trajectories it cannot be saved It is hard to make sense of so much information One way to reduce the displayed trajectories is to filter short trajectories Click on the Filter Options button to filter out trajectories under a given length Enter 144 and click OK All the trajectories will disappear you can also see the message in the results window 0 trajectories remained after filter Since the length of the movie is 144 frames there are no trajectories longer then 144 frames Filter again with O as input All tra
4. H Qian M P Sheetz and E L Elson Single particle tracking analysis of diffusion and flow in two dimensional systems Biophys J 60 4 910 21 Oct 1991 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 1742458 92075891 0006 3495 Journal Article K Ritchie and A Kusumi Single particle tracking image microscopy Meth ods Enzymol 360 618 34 2003 URL nttp www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list_uids 12622171 22509097 0076 6879 Journal Article M Rossner and K M Yamada What s in a picture the temptation of image manipulation J Cell Biol 166 1 11 5 Jul 5 2004 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 15240566 0021 9525 Print News M J Saxton Single particle tracking the distribution of dif fusion coefficients Biophys J 72 4 1744 53 Apr 1997 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 90836078 97237200 0006 3495 Journal Article M J Saxton and K Jacobson Single particle tracking applications to mem brane dynamics Annu Reo Biophys Biomol Struct 26 373 99 1997 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 9241424 97385410 1056 8700 Journal Article R
5. 00000000000000000 00000000000000000 00000000000000000 00000000000000000 0 0 0 0 0 0 0 0 0 000000000000 000000000000 Figure 1 48 Erosion figure taken from DIP 75 EMBL CMCI Image Basic Course 1 3 Filtering Binary Options x Iterations 1 25 d Count 1 8 1 v Black Background OK Cancel Figure 1 49 Setting Iteration 1 3 8 Morphological processing Opening and Closing Combinations of morphological operations can be very useful in remov ing many artifacts present in images This will become very useful after segmenting an image The first operation we will see is opening which is an erosion followed by dilation Opening smooths object contours breaks thin connections and removes thin protrusions After opening all objects smaller than the structuring element will disappear Closing is a dilation followed by erosion Closing smooths object contours joins narrow breaks fills long thin gulfs and fills holes smaller than the structuring element Exercise 1 3 4 1 Load noisy fingerprint tif and broken text tif Apply opening and closing to the images by Process gt Binary gt Open and Process gt Binary gt Close Exercise 1 3 4 2 Optional Next we do morphological processing using anisotropic structuring element Invoke Plugins gt CMCICourseModules gt Morphology and de sign vertical structuring element first with 1 diameter 3 This is
6. Exercise 1 5 5 1 We try measuring the speed of tubulin flux using Kymographs For doing this exercise you need Slice Remover PlugIn 22To install the plugin refer to 1 6 2 If you are using Fiji Slice Remover is already in stalled Access the command by Image Stack Manipulation Slice 118 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series 1 Load the image stack of a spindle labeled with speckle amounts of tubulin Name of the file is control 1 stk Observe the movie Note that the tubulin speckles flux towards the both spindle poles One way to measure the rate of flux is to create a kymograph along a straight line that runs from one spindle pole to the other 2 Remove initial 28 slices by plugins gt course gt Slice Removal Fig 1 87 BETIS x First Slice fi Last Slice 28 Increment OK Cancel Figure 1 87 Slice Removal dialog 3 Contrast enhance and then convert to 8 bit Then do maximum intensity Z projection Image gt Stacks gt Z projection Choose segmented line ROI tool should right click to choose Then trace one of the tracks in the projection image Fig 1 88 Go back to the stack Edit gt Selection gt Restore Selection Then Image gt Stacks gt Reslice Don t change parameters in the dialog window simply OK In the kymograph you just now generated use straight line ROI tool to make a selection along diago nal bright signal
7. Fi p 0 0 0 0 0 0 0 0 0 0 0 0 0 0 coocooocoowwoooocooo coocoocoocowwooococoo coocoocoocoocococomwmwocooo 0 0 0 0 0 0 0 0 0 0 a b Figure 1 3 White line a corresponds to non zero numbers b Note The pixel values are not written to the hard disk as a 2D matrix but as a single array The matrix is only reproduced upon loading according to the width and height of the image Pixel values of image stacks 3D or 4D are also in 1D array in hard disk 3D or 4D matrix is reproduced according 6 EMBL CMCI Image Basic Course 1 1 Basics of Basics to additional information such as slice and channel number Such informa tion is written in the header region of the file which precedes the data region of the file where the 1D array of pixel array is contained The header structure depends on the file format many such formats exist the most common are TIFF or BMP we will see more in file formats and header section 1 1 2 Image Bit Depth Image file has a defined bit depth You might have already heard terms like 8 bit or 16 bit and these are the bit depth 8 bit means that the gray scale of the image has 28 256 steps in other words the grayness between black and white is divided into 256 steps In the same way 16 bit translates to 216 65536 steps hence for 16 bit images one can assign gray level to pixels in a much more precise way in other words grayness resolution
8. automatic classificaiton we will try this technique in a later section The edge detection was already mentioned in the find edge kernel section 1 3 2 Edge detection is one of those feature extractors and is useful for segmenting object edges In the ImageJ menu we could use commands Process gt Find Edges uses Sobel kernel or Plugins gt Feature Extraction gt FeatureJ gt FeatureJ Edg uses gradient filter that do the job but here we try doing it step by step starting from manually preparing Sobel kernels We process an image of microtubule filament Exercise 1 4 2 1 Open image microtubule tif We expect that the values will exceed 255 8 bit so convert the image to 32 bit by Image gt Type gt 32 bit To apply Sobel kernel in x and y direction duplicate the image by Image Duplicate To the original image apply convolution in x axis Do Process gt Filter gt Convolve and input the follow ing kernel 10 1 2 0 2 1 0 1 Be sure to make space between elements Then to the duplicated im age apply convolution with the following kernel 100 EMBL CMCI Image Basic Course 1 4 Segmentation 1 2 1 0 0 0 1 2 1 Now we must get the root square of the sum of squared of each image which is Result microtubule tif microtubule 1 tif For this calculation one could use ImageMath function in Image or ImageExpression Parser in Fiji Depending on your se
9. 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 0 0 0 0 0 oo oc oF j j j jd E pub oA pd jd pad 0 ooo Bm jub em jm pd pd em ao oo 0 0 0 0 jm pud pd jm jm pnd c 0 0 oo C j j de jh jd jd oo 0 mM m 0 0 0 0 0 0 0 0 0 0 0 0 0 When the origin is translated to the e locations the structuring element overlaps 1 valued pixels in the original image 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 Figure 1 47 Dilation figure taken from DIP 74 1 3 Filtering EMBL CMCI Image Basic Course 00000000000000000 00000000000000000 00000000000000000 00000000000000000 00000000000000000 0 0 0 0 0 1 1 I I 1 1 1 0 0 0 0 0 1 00 O00 021 1 1L 1 I 1 0 0 0 0 0 00 07070 11 1 051 1 1 70 070 0 0 00000000000000000 0 000 00 0 070 0 0 0 0 0 0 0 0 00000000000000000 00000000000000000 00000000000000000 Output is zero in these locations because the structuring element overlaps the background n ad Es HET RR nt R2 SRR 00000000000000000 00000000000000000 00000000000000000 oo coo coo co coo oo oc oo oo oo oo oo oo oo oo coo coo 00000000000000000 00500 0 171 11 1 1 1 0 0 0 0 0
10. Likewise images could also be subtracted multiplied and divided by num ber 14 EMBL CMCI Image Basic Course 1 1 Basics of Basics Exercise 1 1 4 1 Simple math using 8 bit image Prepare a new image following the initial part of the exercise 1 1 1 Now bring the mouse pointer over the image and check the value that appears in the status bar in the Image window All pixel values in the image should be value 0 x y in the status bar Commands for mathematical operations in ImageJ are as follows Process gt Math gt Add Process gt Math gt Subtract Process gt Math gt Multiply Process gt Math gt Divide OK Cancel Figure 1 9 Add dialog Add 10 to the image Do Process gt Math gt Add A dialog window pops up and you can for instance input 10 and press OK Now place the mouse pointer over the image to check that the pixel values actually became 10 Then select the pen tool from the tool bar and draw a diagonal line in the window Check again the pixel value The line you just drew has pixel value 255 Then add 10 again to the image by Process gt Math gt Add Check the pixels by placing the pointer The black part is now 20 but what happened to the white In the previous exercise 1 1 1 we converted the image to a text file and then checked the pixel values but it is also possible to check the value pixel by pixel using this method Fro
11. was used to spatially calibrate the image Mean Gray Value Average gray value within the selection This is the sum of the gray values of all the pixels in the selection divided by the number of pixels Reported in calibrated units e g optical density if Analyze gt Calibrate was used to calibrate the image For RGB images the mean is calculated by convert ing each pixel to grayscale using the formula gray red 4 green 4 blue 3 or gray 0 299red 0 587green 0 114blue if Weighted RGB Conversions is checked in Edit gt Options gt Conversions Standard Deviation Standard deviation of the gray values used to generate the mean gray value Uses the Results table heading StdDev Modal Gray Value Most frequently occurring gray value within the selection Corresponds to the highest peak in the histogram Uses the heading Mode Min amp Max Gray Level Minimum and maximum gray values within the selection Centroid The center point of the selection This is the average of the x and y co ordinates of all of the pixels in the image or selection Uses the X and Y headings Center of Mass This is the brightness weighted average of the x and y coordinates all pix els in the image or selection Uses the XM and YM headings These coordi 142 EMBL CMCI Image Basic Course 1 6 Appendices nates are the first order spatial moments Perimeter The length of the outside boundary of the selection Uses the he
12. 3 The sum of multiplication is divided by 9 which is the sum of all elements in the kernel This is to normalize the convolution so that the output value will not to be too large compared to the original We then shift the kernel one step in x direction and apply the kernel for calculating 2 1 position output 2 1 0x1 0x1 0x 14 10x1 10x1 10x1 0x1 0x1 0x1 9 30 9 3 You might have now understood that the smooth operation is actually averaging the values in the surrounding Applying the kernel through the image the new image will be 022 2 0 033 3 0 033 3 0 033 3 0 022 2 0 The vertical line is then now broader and darker the smoothing effect Note that the pixels in the first and the 5th rows were calculated with zero padding so that values are 2 rather than 3 This is the boundary effect unavoidable with any filtering by convolution We could attenuate this effect if we do padding by duplicating neighboring pixels Then the result of convolution then becomes 68 EMBL CMCI Image Basic Course 1 3 Filtering O O O O U 00 C02 C0 Q2 U 00 C02 Q0 CO U 00 Q0 Q0 Q2 co cc o Exercise 1 3 2 1 Working with kernel in Image one can design specific kernels to be used for filtering images by convolution Open microtubule tif image and zoom up so you can see individual pixels Do Process gt Filter gt Convolve A pop up window appears see the image below One could edit the kernel Be
13. EMBL CMCI Image Basic Course 1 3 Filtering 112 11 12421 248 42 12421 112 11 Gauss 7 x 7 Sigma 3 111 2 111 122 4 221 224 8 422 24 8 16 842 224 8 422 122 4 221 111 2 111 Gauss 15 x 15 Sigma 7 22 3 4 5 5 6 6 6 5 5 4 322 23 4 5 7 8 8 8 7 7 5 4 32 34 6 7 9 10 10 11 10 10 9 7 6 43 45 7 9 10 12 13 13 13 12 10 9 7 5 4 5 7 9 11 13 14 15 16 15 14 13 11 9 7 5 5 7 10 12 14 16 17 18 17 16 14 12 10 7 5 6 8 10 13 15 17 19 19 19 17 15 13 10 8 6 6 8 11 13 16 18 19 20 19 18 16 13 11 8 6 6 8 10 13 15 17 19 19 19 17 15 13 10 8 6 5 7 10 12 14 16 17 18 17 16 14 12 10 7 5 5 7 9 11 13 14 15 16 15 14 13 11 9 7 5 45 7 9 10 12 13 13 13 12 10 9 7 5 4 3 4 6 7 9 10 10 11 10 10 9 7 6 4 3 23 4 5 7 8 8 8 7 7 5 4 32 22 3 4 5 5 6 6 6 5 5 4 322 71 EMBL CMCI Image Basic Course 1 3 Filtering Median None linear filters are so called because the result of applying the filter is non linear Median filter used for the removal of noise is one of such filters In ImageJ the command will be Process gt Filter gt Median Following is the principle of how median filter works When we apply median filter with a 3 x 3 size ImageJ samples 3 x 3 neighbors surrounding the target pixel Graphically if the sampled region looks like below target position contains 2 now oO A e O1 N N WwW O Then we align these numbers in the ascending order 123456789 We take the median of this sequence 5 and replace the value 2 with 5 1 3 3 Morpho
14. Fig 1 89 Install macro K_read_kymoLineROL txt by PlugIns gt Macros gt Install and then select the macro file in file chooser and click OK Do PlugIns gt Macros gt Show Line Coordinates and Speed Results will ap pear in the Log window Note 1 Jens Rietdorf and Arne Seitz made a Kymograph plug in for ImageJ It enables multiple ROI selections to make your life easier http www embl de eamnet html kymograph html Remover 119 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series RL o 650x515 pixels 16 bit 652K Figure 1 88 Tracing Projected Image Note Small Yellow Segmented ROI in the middle of the image Reslice of control 1 1600 loj xi 32x10 pixels 16 bit OK Figure 1 89 Measurement of Kymograph 120 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series Note 2 If you want to quantify kymograph with ambiguous patterns you could try using kymoquant an ImageJ macro For more de tails see http cmci embl de downloads kymoquant 1 5 6 Manual Tracking In the simplest case tracking can be done manually The user can read out the coordinate position of the target object directly from the imaging soft ware In Image user can read out position coordinate indicated in the sta tus bar by placing the cross hair pointer over the object Then coordinates can be listed in standard spreadsheet software such as Microsoft Excel for further analysis An
15. Image must be restarted to see the plugin in the menu By default Plug in appears under Plugins If you want to change the location of the plugin in the menu one can change the place by using Plugins Utilities Control Panel If you have too many Plugins there will be significant probability of so called Class conflicts Classes are the modules in Image and Plugins If there are two identical classes in ImageJ then these classes causes the conflict in the process To avoid this one could have multiple Image and install Plugin in each of them for different purposes that there will be no Plugin overloads Fiji Select Plugins gt Install Plugins then choose the plugin file jar or class file 140 EMBL CMCI Image Basic Course 1 6 Appendices 1 6 5 App 1 75 List of accompanying PDF ImageJ_Manual pdf ImageJ Manual written by Tony Collins Cell Imaging Core facil ity Toronto Western Research Intsitute rossner yamada pdf JCB paper about manipulation of image data e TimeSeries Analyzer pdf Manual for the Time Series Analyzer PlugIn Manual Tracking plugin pdf Manual Tracker PlugIn manual 141 EMBL CMCI Image Basic Course 1 6 Appendices 1 6 4 App 2 Measurement Options Copied from http rsb info nih gov ij docs guide userguide 27 html4 toc Subsection 27 7 Area Area of selection in square pixels or in calibrated square units e g mm um etc if Analyze gt Set Scale
16. Tracked data can be saved by activating Result window and File Save as OPTIONAL Copy and paste the result table and plot the track in Excel 1 5 7 Automatic Tracking Automatic tracking reduces the work loads of the researcher enabling them to deal with a huge amount of data Statistical treatments can then be more reliable In addition the results can be regarded more objective than manual tracking Automatic tracking is an optional function that can be added onto some imaging software including Image However these read ily available functions are not adaptable for all analyses since the character istics of target objects in biological research vary greatly Especially when the target object changes shape over time further difficulty arises We try an automatic tracking plugin in this section but keep in mind that the al gorithm for automated tracking has large variety so that knowing the al gorithm well and examination of whether a certain algorithm matches to your object is inevitable for successful tracking 25 Texts of this section is mostly from Miura and Rietdorf 2006 123 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series In the following I will list some of the standard methods for the automatic segmentation of objects which is the first part of the automated tracking Detected objects are then linked from a frame to the other CentroidMethod The centroid is the average of all pixel c
17. Z or T To scroll through X or Y we use Orthognlal View You could view a stack in this mode by Image gt Stack gt Orthogonal View Running this command will open two new windows at the right side and below the stack These two new windows are showing YZ plane and XZ plane Displayed position is indicated by yellow crosses These yellow lines are movable by mouse dragging Exercise 1 1 11 3 Open image mitosis_anaphase_3D tif Run a command Image gt Stack gt Orthogonal View Try scrolling through each axes x y and z 3D viewer Instead of struggling to visualize 3D in 2D planes we could also use the power of 3D graphics to visualize three dimensional image data 3D viewer 45 EMBL CMCI Image Basic Course 1 1 Basics of Basics 7 16 4D_mitosis_256sq166 256 00x25 80x256 pixels 8 bit a XY b YZ 256x80 pixels 8 bit 20K c XZ Figure 1 35 Orthogonal View of a 3D stack is a plugin written by Benne Schimidt It uses Java OpenGL JOGL to ren der three dimensional data as click and rotatable object on your desktop We could try to use the 3D viewer with the data we have been previously dealing with Exercise 1 1 11 4 Open image mitosis anaphase 3D tif Run a command Plugins 3D Viewer A parameter input dialog opens on top of 3Dviewer window Change the following parameters Image Choose mitosis anaphase 3D tif Display as Choose Surface Color Could be any color In
18. brightness and contrast Try changing these values and studying the effect on the image QUESTION What is the problem with the adjustment shown in the right side of the figure below Minimum Minimum Lis es ie Maximum Maximum a Eg vj Brightness Brightness af Contrast Contrast Auto Reset a b Figure 1 42 a Histogram of gel_inv tif before enhancing contrast b Some bad adjust ments example Exercise 1 2 4 2 Continued from above the original image file is not changed at this 60 EMBL CMCI Image Basic Course 1 2 Intensity stage When you push Apply button at the bottom right corner the original image is altered numerically Try set the LUT as you like and push the Apply what happened to the Histogram and the LUT indicator you can always Undo the Apply by Edit gt Undo or revert to the original file by File gt Revert Exercise 1 2 4 3 With RGB image it is possible to adjust the brightness and contrast for individual color channel Open the image RGB cell tif then do Image Adjust Color Balance There is a pull down menu to specify the channel you want to change Try changing different channels to optimize the image 1 25 Image correlation between two images co localization plot In many experiments we need to compare the localization of two or more proteins and examine whether those proteins are co localized In many cases this has been evalu
19. choose the menu item EMBL Samples m51 tif Otherwise if you have sample images saved in your local machine 10 EMBL CMCI Image Basic Course 1 1 Basics of Basics open the image by File gt Open gt m51 tif Choose the line se lection tool and draw a vertical line should be yellow by default called line ROI Then do Analyze gt Plot Profile A win dow pops up See figures 1 4 and 1 5 m5 1 tif 320x510 pixels 8 bit 159K Figure 1 4 Setting a vertical line Roi aixi 4138 8 3 E E o V 213 UT o CT ad rar e a i Distance pixels 487 f st save Copy Figure 1 5 profile of that ROI Figure 1 4 is the profile of the pixel values along the line ROI you just have marked on the image Fig 1 5 X axis is the distance from the starting point in pixel and the y axis is the pixel value along the line ROI The peak corresponds to the bright spot at the center of the 11 EMBL CMCI Image Basic Course 1 1 Basics of Basics image Let s convert the image to 8 bit First check the state of Conver sion option by Edit gt Option gt Conversion Make sure to tick Do Image gt Type gt 8 bit The line ROI is still there after the conversion Do Analyze gt Plot Profile again You will find another graph pops up Compare the previous profile 16 bit and the new profile 8 bit Conversion modifies the y value Shapes of the profile look
20. target particles must be segmented by image threshold Try thresholding the image What you would find out immediately is that there is shading in the image that rice can not be uniformly se lected by thresholded For this reason do the background subtraction as we already did in 1 3 5 We then work with the subtracted image in the following Set Threshold again and then click Apply in the Threshold dialog window The image then should be black and white This binarized segmented image will be used as a reference for setting boundary of each rice grain Open file rice tif again Since there is already a window with same title this second one should have a title rice 1 tif By having this original image boundary that will be set using binarized image will 106 EMBL CMCI Image Basic Course 1 4 Segmentation be redirected when measurement is actually taking place Before doing particle analysis you must specify what you want to measure for each particle through Analysis gt Set Measurements Select Area Circularity Centroid and Perimeter details on these pa rameters are written in Appendix 1 6 4 In addition set redirect to to the original image rice 1 tif so that the measurement will be done with the original image and not with the thresholded image Activate the threshold the image and then do Analyze Analyze Particles A dialog window pops up Input following param eters e Size 0
21. 153 989532 132 864670 4 795121 3 200076 8 058751 8 150 077560 135 210281 6 471988 3 328422 7 164520 9 150 077560 135 210281 6 471988 3 328422 7 164520 10 150 231461 136 617905 9 787056 3 064149 26 286119 11 150 231461 136 617905 9 787056 3 064149 26 286119 12 149 775223 137 932922 9 405724 3 051208 22 606100 13 149 796631 137 935913 9 499812 3 075651 23 014484 14 148 993317 137 686401 6 237571 3 227575 11 401782 15 148 993317 137 686401 6 237571 3 227575 11 401782 16 148 938690 138 170959 6 908313 3 284760 1 390782 4 All Trajectories Visualize All Traiectories Focus on Selected Traiectorv Click on the Focus on Selected Trajectory button a new window with a focused view of this trajectory is displayed This view can be saved with the trajectory animation through the File menu of Image Look at the focused view and compare it to the overview window in the focused view the white trajectory that is close to the yellow is not displayed __ Trajectory number 32 Bele All Trajectories Visual BEIE The particle is displayed but the white trajectory animation is not This is because we selected Focus on Selected Trajectory Close this focus view Now we what to focus on area for number of trajectories view we will focus on the area of the yellow and white trajectories as shown here Select a rectangle region of interest around these trajectories click and drag the mouse on the overview to include them Click on
22. Basic Course 1 5 Analysis of Time Series et al 2001 Another advantage of the Gaussian fitting method is that the results are in sub pixel resolution Such high resolution can for example enable the analysis of molecular diffusion in the nm resolution More dis cussion on the localization accuracy in the position measurements can be found in the literature Ober et al 2004 Martin et al 2002 Thompson et al 2002 Pattern Matching Method In this method a kernel containing a template pattern of the object is com pared to different positions within the image in order to find a position with the highest similarity to the kernel A cross correlation function C x y is usually used to evaluate the resemblance of the template with different parts of the image Gilles et al 1988 n 1m 1 C xy 9 Vo Y Ix y D KGj K 1 14 i 0 j 0 where I x y is the intensity distribution of the image frame and K i j isa n x m pixels kernel that contains the template image pattern K is the mean intensity of the kernel C x y will be maximal at the position where the pattern best matches In actual application the template pattern is sampled in the k th image frame I x y by the user manually or by semi automatic segmentation Then the cross correlation value C x y between this tem plate kernel and the consecutive k 1 th image frame Ij 4 x y can be cal culated Gelles et al 1988 introduced the following formula for obtai
23. ImageJ plug in is also freely available to assist such simple way of track ing An obvious disadvantage of the manual tracking is that the mouse clicking by the user could be erroneous as we are still human who gets tired after thousands of clicking For such errors measurement errors can be estimated by tracking the same object several times and this error could then be indicated together with the results Otherwise automated tracking is more objective but if you have only limited number of tracks to analyze manual tracking is a best choice before start to explore complex parameter space of automated tracking setting The manual tracking can be assisted by an ImageJ plug in Manual Track ing This plug in enables the user to record x y coordinates of the posi tion where the user clicked using mouse in each frame within a stack The download site has a detailed instruction on how to use htcpi y rsb lnfo nilh gov ldg pluglins track rrack html Exercise 1 5 6 1 Manual Tracking In a spindle microtubules attach to chromosomes through structures called kinetochores In the stack kin stk kineto chores are labeled with a fluorescent marker Your task now is to track the movement of individual kinetochores using the ManualTracker PlugIn Enhance contrast and convert the image stack from 16 bit to 8 bit just to decrease the memory load If your computer is powerful 121 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series eno
24. ParticleTracker and the right display and analysis options upon completion of it Written by Guy Levy at the Computational Biophysics Lab ETH Zurich To use this tutorial you should already be familiar with ImageJ and have the plugin installed It is also recommended to read the user manual before starting In this tutorial we will guide you through an example using movies sample data e TransportOfEndosomalVirus zip a real experimental movie image sequence from Suomalainen et al J Cell Biol 144 4 657 672 Kindly provided by The Cell Biology group Institute of Zoology University of Zurich Fast minus and plus end directed transport of endosomal ts1 virus labeled with Texas red Virus was added to TC7 MAP4 MTB GFP cells and 72 TR images were recorded with intervals of 1 5 s starting 20 min p i The GFP signal is indicated at the beginning Bar 10 um e Artificial zip Artificial image sequence will be used later in the tutorial for a specific example First Step download sample movie and open it Second Step select particle detection parameters and preview Third Step viewing the results Fourth Step re linking the particles First Step download sample movie and open it Download the TransportOfEndosomalVirus zip file Both the movies are actually sequence of separate image files tif all in a zipped folder After downloading the zip file unzip it to your preferred folder and Start Image Load the
25. Pixel value does not change by this operation but the pixel intensity changes that the image ap pears differently now Check the LUT again by do Image gt Color gt Show LUT Actual numbers in each LUT could be checked using list button at the left bottom corner of each LUT window ioj xl j iojx Cell Colony jpg x Cell Colony jpg o 406x408 pixels 8 bit 161K 406x408 pixels 8 bit 161K a b Figure 1 22 Grayscale LUT a converted to spectrum LUT b 26 EMBL CMCI Image Basic Course 1 1 Basics of Basics Look Up Table 21 x lolx 326x188 pixels RGB 236K 326x188 pixels RGB 236K 255 255 a b Figure 1 23 a Grayscale LUT and b spectrum LUT 1 1 8 Image File Formats In this part I will discuss on issues related to image file formats including e Header and Data Data Compression Image file contains two types of information One is the image data the matrix of numerical values Another is called header which contains the information about the architecture of image Since software must know information such as bit depth width and height and the size of the header before opening the image the information is written in the header The total file size of an image file is then Total filesize headersize datasize There are many different types of image formats such as TIFF BMP PICT and so on The organization of the information in the header is specific t
26. Series A maximum intensity projection of an entire image stack is given by M x y max Ij x y 1 11 The maximum on the right hand side is the maximum in intensity value at a given pixel position over all time frames in an image Maximum pro jections collapse the entire dynamics of the stack onto a single plane This function is especially useful for visualizing entire trajectories of moving particles on a single plane Additional merit is that this operation discards noise within the sequence Exercise 1 5 2 1 Open image listeriacells stk Then do all types of projections you could choose by Image gt Stack gt Z projection 116 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series Question Which is the best method for leaving the bacteria tracks 15 3 Measurement of Intensity dynamics Temporal changes in fluorescence level matters a lot in biological exper iments since the change is directly related to the molecular mobility and production Here we study how to obtain the intensity dynamics out of image sequences Exercise 1 5 3 1 Load image stack 1703 2 3s 20s stk This is a sequence of FRAP experiment Draw a ROI could be any closed ROI surrounding the area where the photobleaching takes place Then do Image gt Stacks Plot z axis Profile There will be two windows Results window and a graph of inten sity dynamics Try adding Second ROI to measure the background simply by mak i
27. The higher the number in the cutoff field the more suspicious the algorithm is of false particles This could be very helpful when one understand the method for non particles discrimination as described in the original algorithm It can also lead to real particles discrimination when the value is too high After setting the parameters for the detection we will go with radius 3 cutoff 0 percentile 0 6 you should set the particle linking parameters The parameters relevant for linking are e Displacement The maximum number of pixels a particle is allowed to move between two succeeding frames e Link Range The number of subsequent frames that is taken into account to determine the optimal correspondence matching These parameters can also be very different from one movie to the other and can also be modified after viewing the initial results Generally in a movie where particles travel long distance from one frame to the other a large link range should be entered In this movie a link range of 20 is appropriate Set it to 20 The linkrange value is harder to set before viewing some initial results since it is mainly designed to overcome temporary occlusion as well as particle appearance and disappearance from the image region and it is hard to notice such things at this stage Still an initial value has to be set the default is 2 but we will continue with 3 We will return to these parameters later with a different movie
28. Value Min amp Max Gray Value T Centroid Center of Mass Perimeter Bounding Rectangle J FitEllipse Feret s Diameter Iv Integrated Density Median Skewness Kurtosis Area Fraction Stack Position iv Limitto Threshold MM Display Label Invert Y Coordinates Scientific Notation Redirect To None Decimal Places 0 9 5 OK Cancel Figure 1 39 Set Measurement Window Results File Edit Font Area Mean Innen 1 2577 45 158323632 116373 a b Figure 1 40 a Tracing Cell Edge by Segmented ROI and Measuring the Intensity within selected area 58 EMBL CMCI Image Basic Course 1 2 Intensity background intensity by creating a new ROI in the area where there is no cell Figure 1 41 Measurement of Background NOTE When no ROI is active no yellow bounding rectangle or so then the measurement is performed on the whole image 1 24 Image transformation Enhancing Contrast Some of you might already have experience with the contrast enhancing of digital images since in most of imaging software like the ones that come with digital camera usually have this function Low contrast images have small difference in the tones and the objects are difficult to observe If the contrast is too high then the tone difference is so much that the picture is over exposed Adjustment of contrast controls the tone difference to op timize the visual resolution The contrast
29. algorithm could be found at Process gt Binary gt Watershed To understand the watershed transform we view a grayscale image as a topological surface where the values of f x y correspond to heights ab Watershed ridge line FIGURE 10 18 is a Gray scale image of dark blobs b Image viewed as a surface with labeled watershed ridge line and catchment basins a b Catchment basins Figure 1 78 Watershed Principle Consider the topographic surface on the right Water would collect in one 103 EMBL CMCI Image Basic Course 1 4 Segmentation of the two catchment basins Water falling on the watershed ridge line separating the two basins would be equally likely to collect into either of the two catchment basins Watershed algorithms then find the catchment basins and the ridge lines in an image Fig 1 78 1 80 The algorithm works as follows Suppose a hole is punched at each re gional local minimum and the entire topography is flooded from below by letting the water rise through the holes at a uniform rate Pixels below the water level at a given time are marked as flooded When we raise the water level incrementally the flooded regions will grow in size Eventu ally the water will rise to a level where two flooded regions from separate catchment basins will merge When this occurs the algorithm constructs a one pixel thick dam that separates the two regions The flooding continues until the entire
30. background color is black Process gt FFT gt Invert FFT to see the spatial domain image after filtering Similar to this exercise we could separate a spatial domain image to high frequency part and low frequency part such as shown in an example be low Adding high frequency Fig 1 68a and low frequency Fig 1 68b stripe images results in an overlapped image of two frequencies Fig 1 68c This image math could be done easily using Process gt Image Math From this overlapped image it is not easy to isolate two original images by spatial domain filtering but it could be done in a simple way with the frequency 91 EMBL CMCI Image Basic Course 1 3 Filtering domain image Fig 1 69a FFT image could be separated in to two images one from the center called low pass Fig 1 69b and the other from pe ripheral called high pass Fig 1 69c Process gt FFT gt Invert FFT of each FFT image would result in the low frequency stripe image Fig 1 70b and the high frequency stripe image Fig 1 71b EES olx 50x50 pixels 8 bit 50x50 pixels 8 bit 50x50 HIT 8 bit ra ZZ Figure 1 68 Images of a high frequency pattern b low frequency pattern c a and b combined EES oxi ENGEL inixi mee unii 64x64 pixels 8 bit 4K fami nx 64x64 pixels 8 bit 4K 64x64 pixels 8 bit 4K b o Figure 1 69 a FFT image 2D power spectrum of Fig 1 68c could be sepa
31. can do both spatial domain convolu tion and frequency domain convolution We first do the convolution in spatial domain similar to what we have done already in the section 1 3 1 We do this again for a com parison with convolution in frequency domain Spatial Domain Convolution Process gt Filter gt Convolution In the convolution panel click open and choose the Laplacian you created in above Be sure to check Normalized Result should look like 88 EMBL CMCI Image Basic Course 1 3 Filtering 256x256 pixels 32 bit inverting LUT 256K Figure 1 63 blobs gif Laplacian kernel applied Frequency Domain Convolution We first prepare a Laplacian kernel image Open the kernel lapla cian3x3 txt by Image gt Import gt Text Image It should be very small but if you zoom up the image it should look like 3x3 pixels 32 bit OK Figure 1 64 Laplacian3x3 txt text image zoomed up Then adjust the image size to 256 x 256 Image gt Adjust gt Canvas Size Be sure to set position to center and check zero fill Then we do the convolution by Process gt FFT gt FD math 89 EMBL CMCI Image Basic Course 1 3 Filtering laplase3x3 txt A ol 256x256 pixels 32 bit 256K Figure 1 65 Laplacian3x3 txt canvas size adjusted Choose blob gif and laplace3x3 txt for imagel and image2 respec tively Operation should be Convolve Uncheck Do Inverse
32. certain pixel value Intensity his togram of an image is a plot that shows distribution of pixel counts over a range of pixel values typically its bit depth or between the minimum and the maximum pixel value with in the image Histogram is useful for exam ining the signal and background pixel values for determining a threshold value to do segmentation We will study image threshold in 1 4 Exercise 1 2 1 1 Open Cell Colony tif Do Analyze gt Histogram Anew window appears The x axis of the graph is pixel value Since the image bit depth is 8 bit the pixel values ranges from 0 to 255 The y axis is the number of pixels so the unit is count Since 255 white and 0 black the histogram has long tail towards the lower pixel value This reflects the fact that the image has white background and black dots Check pixel values in the histogram by placing the mouse pointer over the plot and move it along the x axis Pixel count appears at the bottom of the histogram window Switch to the cell colony image and place the pointer over dark spot Read the pixel value there and then try finding out the number of pixels with the same value in the histogram What is the range of pixel values which corresponds to the spot signal 52 EMBL CMCI Image Basic Course 1 2 Intensity Histogram could also be used for enhancing contrast of image Several different algorithms are available for this histogram normalization his togram equali
33. do not use correlation in this course but correlation is very much similar to convolution In case of convolution the 1D kernel was rotated by 180 degrees In correlation no rotation is done and used as it is and the rest of the algorithm is same Correlation is often used for pattern matching to find out the pixel intensity distribution in an image using a template image as a kernel For example many object tracking algorithm utilizes correlation for searching target object from one frame to the other In ImageJ PIV particle image velocimetry plugin uses the correlation to estimate motion vector field Convolution and correlation Two closely related bilinear operations that are especially im portant for information processing are convolution and correlation In the simplest case correlation can be described as a compari son of two fields at all possible relative positions More specifi cally if x is the correlation of two one dimensional fields and UV X y then x r reflects how well p and y match in an inner product sense when relatively displaced by r Mathe matically x r f es nouis Higher dimensional correlations are the same except that r is a relative displacement vector rather than a scalar https sites google com site qingzongtseng piv 65 EMBL CMCI Image Basic Course 1 3 Filtering Convolution x wp is essentially the same as correlation except that the field is reflected be
34. drive membrane gly coprotein movements Nature 340 6231 284 8 Jul 27 1989 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 2747796 89314153 0028 0836 Journal Article J Suh M Dawson and J Hanes Real time multiple particle tracking applications to drug and gene delivery Adv Drug Deliv Rev 57 1 63 78 Jan 2 2005 URL nttp www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 15518921 0169 409x Journal Article Review C Tardin L Cognet C Bats B Lounis and D Choquet Di rect imaging of lateral movements of ampa receptors in side synapses Embo J 22 18 4656 65 Sep 15 2003 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 12970178 22850113 0261 4189 Journal Article R E Thompson D R Larson and W W Webb Precise nanometer localiza tion analysis for individual fluorescent probes Biophys J 82 5 2775 83 May 2002 URL http www ncbi nlm nih gov entrez query 137 EMBL CMCI Image Basic Course REFERENCES fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 11964263 21961515 0006 3495 Journal Article W Wallace L H Schaefer and J R Swedlow A workingperson s guide to de convolution in light microscopy BioTechniques 31 5 1076 8 1080 1082 passim November 2001 ISSN 0736 6205 URL h
35. enhancement primarily changes the LUT so that the original image data is unaffected you will see in the following exercise The original data is changed only when you click Ap ply button Then the pixel values are re scaled according to the LUT you set Care must be taken for contrast enhancement since pixel values are altered This could be regarded as fraud or manipulation in science especially if you are measuring image intensity If all images that you are trying to compare were equally contrast enhanced then the images could eventually be compared Even then there will be another pit fall if you artificially 59 EMBL CMCI Image Basic Course 1 2 Intensity saturate the image this happens often especially in the case of low bit depth images Exercise 1 2 4 1 Open image gel_inv tif Do Image gt Adjust gt Brightness Contrast Pop up window appears which looks like the figure below left The graph shown in the upper part of the window is the histogram of the image just like you have seen in section 1 2 1 Histogram Since the im age is 8 bit scale in the x axis is 0 to 255 There is also a black diagonal line This corresponds to the LUT of the image on the screen pixel values in x is translated into the gray value on the screen brightness on the screen is y axis The slope and position of the LUT can be altered by click and drag ging four sliding bars under the graph each with the name minimum maximum
36. image is segmented into separate catchment basins divided by watershed ridge lines Exercise 1 4 3 1 Open binary image Circles tif You see two circles fused Such sit uation often occurs with cells or organelle and you might want to simply separate them as different objects Figure 1 79 Circles tif Do Process gt Binary gt Watershed You now see that two cir cles are separated at the constricted part of the fused circles If you want to know a bit more details read the following quote how Im ageJ does this Watershed segmentation is a way of automatically separating or cutting apart particles that touch It first calculates the Eu clidean distance map EDM and finds the ultimate eroded points 104 EMBL CMCI Image Basic Course 1 4 Segmentation UEPs It then dilates each of the UEPs the peaks or local max ima of the EDM as far as possible either until the edge of the particle is reached or the edge of the region of another grow ing UEP Watershed segmentation works best for smooth con vex objects that don t overlap too much quote from Image web site Thresholded Cells EDM and UEPs After Watershed Segmentation Figure 1 80 Watershed Segmentation Image first computes the distance transform on binary image The distance transform of a binary image is the distance from every pixel to the nearest nonzero valued pixel as example in Fig 1 81 shows ban o 1 00 1 00 1 41 Fi
37. interpolation algorithm samples pixel val ues in the surrounding of the insertion point and calculates the pixel value for that position One must keep in mind that the result of enlarging or shrinking of image depends on the interpolation method and scientific re sults could be altered depending on the method you use In a more general context this problem is treated as sampling theory With this keyword search more explanation in information theory textbooks and in the Inter net 10 For more details on bilinear interpolation refer to http www cambridgeincolour com tutorials image interpolation htm 49 EMBL CMCI Image Basic Course 1 1 Basics of Basics 11 13 ASSIGNMENTS Assignment 1 1 1 Digital image matrix of numbers Edit a text image using any text editor Be sure to insert space between numbers as separator Save the text file and open it as an image in ImageJ by importing text image function Assignments 1 1 2 bit depth 1 How many gray scale steps does a 12 bit image have 2 Describe in text how a 1 bit image looks like Assignment 1 1 3 bit depth conversion Use m51 tif 16 bit sample image to draw a plot profile as we did in the course In the profile plot window a list button is at the left bottom corner Click the button You will then see a new window containing a column of numbers These numbers can be copy amp pasted to spread sheet software such as LibreOffice or MS Excel or
38. is higher Microscope images are generated mostly by CCD camera or something similar CCD chip has a matrix of sensors Each sensor receives photons and converts the number of photons to a value for a pixel at the corresponding position within the image Larger bit depth enables more detailed conversion of signal intensity infinite steps to pixel values limited steps Why do we use 2 This is because computers code the information with binary numbers In binary the elementary units are called bits and there only possible values are 0 and 1 in the decimal system these units are called digits and can take values from 0 to 9 Coding values with 8 bit means for example that this value is represented as a 8 bit number something like 00001010 10 in decimal Then the minimum value is 00000000 0 in decimal and the maximum is 11111111 255 in decimal 8 bit im age allows 256 scales for the grayness using calculator application in your computer you could easily convert binary number into normal decimals and vice versa In case of 16 bit image the scale is 216 so there are 65536 steps EMBL CMCI Image Basic Course 1 1 Basics of Basics We must not forget that the nature is continuous In conventional math ematics as you learn in school a decimal point enables you to represent infinite steps between 0 and 1 But by digitizing the nature we lose the abil ity to encode infinite steps such that 0 44 might be roun
39. mostly similar so if you normalize two images the curve may overlap This is because the image is scaled according to the following formula he x y min Be x y max hs y min hsQuy lg x y where le x y 16 bit image min Ie x y the minimum value of 16 bit image max Ig x y the maximum value of 16 bit image Ig x y 8 bit image Save the line ROI you created by pressing t or clicking on add in the ROI manager you can open the manager with Analyze Tools ROI manager A small dialog window pops up so click Add button in the right side The numbers in the left column are names of the ROIs you add to the manager they correspond to the coordinates of the start end points of the ROI Now change the option in Edit Option Conversion by un ticking scale when converting Open the 16 bit image again by File Open m51 tif Thenagain do Image Type 8 bit An apparent difference you can observe is that now the picture looks like a overexposed image Find the ROI manager window and click the ROI number you stored in above Same line ROI will appear in the new window Then do Analyze Plot Profile This third profile has a very different shape compared to the previous ones This 12 EMBL CMCI Image Basic Course 1 1 Basics of Basics Conversion Options xj Figure 1 6 Conversion Option Scaling is turned off in this case is because the values
40. of an image The first derivative of choice in image processing is the gradient defined as the vector grad f GG where Gy df dx and Gy df dy are the partial derivatives in the horizontal and vertical directions of the image The mag nitude of this vector is lgrad f G3 G3 The gradient vector points in the direction of steepest ascent The angle of steepest ascent is given by a x y tan 1 G Gy We can estimate the derivatives Gy and Gy digitally by linearly filtering the image with the following 3 by 3 kernels 147 EMBL CMCI Image Basic Course 1 6 Appendices Sobel Prewitt Roberts The Prewitt and Sobel operators are among the most used in practice for computing digital gradients The Prewitt masks are simpler to implement than the Sobel masks but the latter have slightly superior noise suppression characteristics 148 EMBL CMCI Image Basic Course 1 6 Appendices 1 6 6 App 4 Particle Tracker manual Tutorial from http weeman inf ethz ch particletracker tutorial html PDF inserted from next page 149 ImageJ Image Processing and Analysis in Java ParticleTracker Tutorial The ParticleTracker is an Image Plugin for multiple particle detection and tracking from digital videos Sample Data and Tutorial This tutorial provides a basic introduction on how to select the right parameters for the algorithm of
41. parameters to what you think is optimum Set Linking parameters Two parameters for linking detected dots should be set Link Range could be more than 1 if you want to link dots that disappears and reappears If not set the value to 1 Displacement expected maximum distance that dots could move from one frame to the next Unit is in pixels After parameters are set click OK Tracking starts Inspect the Tracking Results When tracking is done a new window titled Results appears At the bottom of the window there are many buttons Click Visualize all trajectories and then a duplicate of the image stack overlaid with trajectories will appear Select a region within the stack using rectangular ROI tool and then click Focus on Area This will create another image stack with only 128 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series that region Since this image is zoomed you could carefully check if the tracking was successful or not If you think the tracking was not successful then you should reset all the parameters and do the tracking again Export the tracking results To analyze the results in other software data should be saved as a file To do so first click All Trajectories to Table Results table will then created In this results table window select File gt Save As and save the file on your desktop By default file type extension is xls excel format but chang
42. processing protocol you can easily apply it to many files automatically image by image Such task is called batch processing Prerequisite for using batch processing function in Image is that all files that you want to process are stored in a single folder Another preparation for batch processing is that you should record the processing command in the following way shown in the exercise Exercise 1 3 8 1 Here we use numbered tiff file series as an example to do batch pro cessing Open sample images spindle frames eg5 spindle 500016 tif In order to to record the processing command do P1ugins Macros Recorder A recorder window pops up igixi Name Macro ijm Figure 1 52 Macro Recorder Window Then go back to the spindle image activate the image window by clicking the title bar and then do Process gt Subtract gt Background Try to input some values in the subtract background dialog click OK and check that the image is background subtracted Fig 1 53 Check the Recorder window again You will see that a new text line is added This text should be like it is shown in Fig 1 54 run Subtract background rolling 5 disable The number after rolling should be adapted to your input some 81 EMBL CMCI Image Basic Course 1 3 Filtering iol x iol xl 391x291 pixels 16 bit 222K 391x291 pixels 16 bit 222K a b Figure 1 53 Spindle images a before and b after the backg
43. put folder In above exercise we had only one processing command but you could add many more text commands which you could extract by using Recorder 1 3 9 Fast Fourier Transform FFT of Image FFT converts a spatial domain image data what you are normally seeing to a spatial frequency domain data FFT is used because 1 In some occasion calculation cab be done much faster in frequency domain than in spatial domain i e convolution involving large ker nels 83 EMBL CMCI Image Basic Course 1 3 Filtering 2 Some image processing techniques can only be performed in the fre quency domain Exercise 1 3 9 1 Reversibility of FFT Open microtubule tif by File gt Open Then apply FFT by Process gt FFT gt FFT A new window showing frequency domain image 2D power spectrum log display appears To check that FFT is re versible apply Process gt FFT gt Inverse FFT Fig 1 56 FFT of microtubule tif 256x256 pixels 8 bit 64K Toeto c Figure 1 56 Reversibility of FFT a Original b FFT c Inverse FFT Here is an intuitive explanation of what frequency domain image is Ori entation of patterns in spatial domain image has a clear relationship with the resulting FFT image We take as example four images with differently oriented stripes vertical diagonal right to left or left to right and horizon tal Fig 1 57 5 When these images are transformed by FFT the re
44. second frame or slices to reduce the stack size by a half To split two channel time series to individual time points without splitting the channels You need to insert a small inset within a stack For such demands tools for stack editing are available under the menu tree Image gt Stack gt Tools Figure 1 28 and figure 1 30 schematically shows what each of these commands does Exercise 1 1 9 3 Creating a Montage Create a new image File gt New gt Image with the following properties Name could be any thing Type 8 bit e Fill with Black e Width 200 34 EMBL CMCI Image Basic Course 1 1 Basics of Basics Stack Slice Manipulation Image Stacks Tools Combine E Concatenate Combine Deinterleave Montage to 2 Make Substack Groupes Z Project Remove Sice Labels Start Aremabon Stop Animabon Amas Sama Combine vertically Ske Remover Suce Keeper Ness Splitter Stack Sorter ex x4 rteriene Destination Lm Stack Inserter X y coordinates Figure 1 28 Editing Stacks 1 These commands are under Image gt Stack gt Tools Courtesy of Sebasti n Tosi IRB Barcelona 35 EMBL CMCI Image Basic Course 1 1 Basics of Basics Stack Slice Manipulation Image Stacks Tools Stack Reverser Images to Stack we Stack to Make Substack Images Ex 1 3 6 8 Start Animation N Stop Animabon Animation Options Stac
45. sequence as animation This com mand can be found at Image gt Stacks gt Tools gt Start Animation Try changing the playback speed by Animation Options This command pops up a dialog to set the playback speed In the main menu this same command is at Image gt Stack gt Tools gt Animation options To exclusively work on stacks there is an icon at the right most position in the ImageJ menu bar Click and then from drop down menu that appeared select Stack tools Video player like interface appears in the tool bar 1 26 Try different buttons to see what action 32 EMBL CMCI Image Basic Course 1 1 Basics of Basics 900 Fiji ORJOLI AATA aA OO lt s gt Add Slice Figure 1 26 Image tool bar in Stack Tool mode Speed 0 1 1000 fps First Frame Last Frame 1 Loop Back and Forth Vf Start Animation Cancel J OK jJ Figure 1 27 Animation Option Window 33 EMBL CMCI Image Basic Course 1 1 Basics of Basics they perform Editing Stacks In many occasions you might need to edit stacks such as Truncate a stack because you only need to analyze a part of the se quence For a presentation you need to combine two different stacks Need to add a stack at the end of another stack To make a combined stack by attaching each frame in a stack by a frame of another stack Typically you want to have two channels of image side by side You need only every
46. sure to make spaces between numbers By clicking OK the kernel will be applied to the image Try replacing the default kernel with the smoothening kernel we stud ied above Apply the kernel to sample image microtubule tif In crease the dimension to 5 x 5 9 x 9 and do the smoothening Question what happened when the size of the kernel became larger Check the preview option so that the change in the kernel could be visualized directly while editing E 111 111 111 Open Save Normalize Kernel Preview OK Cancel Figure 1 46 Convolver Window 69 EMBL CMCI Image Basic Course 1 3 Filtering Sharpen 1 1 1 12 1 1 1 1 This kernel sharpens the image also known as Laplacian Side effect noise is also enhanced Find Edge gradient The following two kernels are applied independently Square root of the sum of the square of two result images will be calculated called Sobel Filter for more details see Appendix 1 6 5 1 2 1 0 0 0 1 2 1 and 10 1 20 2 10 1 Gaussian Blur This kernel blurs in positive sense we call it smooth the image by con volution using a square Gaussian bell shaped kernel The width of the kernel in pixels is 2 sigma 1 where sigma is entered into a dialog box Im ageJ documentation The following Gaussian kernels are respectively ob tained for sigma equal 2 5x5 3 7x7 and 7 15x15 Gauss 5 x 5 Sigma 2 70
47. the Focus On Area button a new window with a focused view of these trajectories is displayed This time the animation of both trajectories is displayed Generally any unfiltered trajectory part that is in the selected area will be shown You may notice that some particles are showing but their trajectory is not animated this is because they are filtered remember we filtered for longer then 100 Close the focus window and reset the filter You can do that by closing the overview window and reopening it by clicking the Visualize all Trajectories button or you can click the filter button and set the min length to O default The last option is better since this way your area selection will stay Click again on the Focus on Area button now all trajectories within the selection area is displayed The size of the focus window for specific trajectory and area focus is determined by the magnification factor relative to the original movie window size Select the pink trajectory the one shown here The trajectory number is 44 amp All Trajectories Visual Sele 1 144 tutorial0000 228x275 pixels 8 hit 8 Focus on Selected Trajectory ui eet oe eet tee oe Filter Options Notice that the rectangle surrounding the selected trajectory is fairly big If we focus on this trajectory with the default magnification factor 6 a large window will be created and may cause memory problems especially in Mac Os For this reason an
48. the dialog window when you execute this operation you will be asked for the Rolling ball radius This should be at least as large as the radius of the largest object in the image that is not part of the background Exercise 1 3 6 1 Open rice tif Do background subtraction by Process gt Subtract gt Background Change the Rolling ball radius and study the effect Question what happens when the rolling ball radius is smaller M4Stanley Sternberg s article Biomedical Image Processing IEEE Computer January 1983 78 EMBL CMCI Image Basic Course 1 3 Filtering Figure 1 51 Background Subtraction from Image site There are several other background subtraction methods that do not de pend on morphological filtering 1 High pass filter 2 Flat field correction by polynomial fitting 3 Deconvolution Pseudo high pass filtering could be done by subtracting largely blurred image by Gaussian blur such as with sigma of 20 from the original im age The same processing could also be be achieved by a single command Process gt Filter gt Unsharp Mask The band pass function under Process gt FFT could be used for the true high pass filtering but in my experience the result of pseudo highpass filtering is more usable for seg mentation purpose For polynomial fitting ImageJ plugin Polynomial_Fit written by Bob Dougherty could be used to estimate the background P Practical information on bac
49. the example shown in fig 1 36 white was chosen Threshold Default value 50 should work OK but you could also try changing it to greater or smaller values This threshold value determines the surface Pixel intensity greater than this value will be considered as object else background e Resampling factor default value 2 should be sufficient If you change this value to 1 then it takes longer time for rendering 46 EMBL CMCI Image Basic Course 1 1 Basics of Basics In case of our example image which is small difference in the rendering time should be not really recognizable After setting these values clicking OK will render the image in the 3Dviewer window Try to click and rotate the object J3t File Edit View Add Landmarks Help eoo Image 3D Viewer File Edit View Add Landmarks Help Add Image mitosis anaphase 3D tif Name mitosis anapha Display as Surface Color White Threshold 50 Resampling factor 2 Channels amp red amp green amp blue Start at time point 0 Cancel OK a Parameter Input Dialog b Surface rendered 3D stack Figure 1 36 Surface rendering by 3DViewer More advanced usages are available such as visualizing two channels or showing 3D time series as a time series of 3D graphics or saving movies For such usages consult the tutorial in the 3DViewer website http 3dviewer neurofly de 11 12 Resampling images Shrinking and E
50. this option to analyze a dot blot assay Median The median value of the pixels in the image or selection Skewness The third order moment about the mean The documentation for the Mo ment Calculator plugin explains how to interpret spatial moments Uses the heading Skew Kurtosis The fourth order moment about the mean Uses the heading Kurt Area Fraction For thresholded images is the percentage of pixels in the image or selection that have been highlighted in red using Image gt Adjust gt Threshold T 144 EMBL CMCI Image Basic Course 1 6 Appendices For non thresholded images is the percentage of non zero pixels Uses the heading Area Stack Position The position slice channel and frame in the stack or hyperstack of the selection Uses the headings Slice Ch and Frame n b For line selections the heading Length is created For straight line selections Angle is recorded even if Fit Ellipse is unchecked Also note that measurements that do not apply to certain selection types may be listed as NaN Infinity or Infinity The second part of the dialog controls measurement settings Limit to Threshold If checked only thresholded pixels are included in measurement calcula tions Use Image gt Adjust gt Threshold T to set the threshold limits This setting affects only thresholded images see Settings and Preferences Display Label If checked the image name and slice number for stacks
51. trans form and OK Image named Result appears Result 256x256 pixels 32 bit Figure 1 66 Frequency domain convolution of Blob image Then do the inverse FFT by Process gt FFT gt Inverse FFT Check that you could get the original image by deconvolve opera tion using FD math 1 3 11 Frequency domain Filtering Filtering using FFT image 2D power spectrum is a way of improving the image quality and also to allow a better segmentation One typical exam 90 EMBL CMCI Image Basic Course 1 3 Filtering Inverse FFT of Result B iri x 256x256 pixels 32 bit 256K Figure 1 67 Frequency domain convolution of Blob image now in Spatial domain ple is to reduce noise As we have seen in Fig 1 61 noise produces high frequency isotropic signal that appears in the periphery of FFT image We could remove noise by simply throwing away peripheral signals in FFT image to reduce noise in the original image Exercise 1 3 11 1 Removing noise using FFT image Open microtubule tif by File gt Open Then apply FFT by Process gt FFT gt FFT A new window showing microtubule converted to a frequency domain image 2D power spectrum appears Make a rectangular ROI covering vertical streak at the center of the FFT im age and then Image gt Clear gt Outside to convert peripheral signals to 0 black If Clear Outside produced white periphery check Edit gt Options gt Colors to see if the
52. 2 Bdsite Of Basics amp 2 2 9 mec Arr DR OR eee p ARR 11 1 Digital image is a matrixofnumbers 1 1 2 Image Bit Depth i odo 9 ww 1 13 Converting the bit depth ofanimage 1 1 4 Math functions Re EX PALESTRA 1 1 5 Image Math o eec ceri caca eaea taeu SS PEG RGB AAG e Anaa ae geek Go to ht te 1 1 7 Look Up Table amp eat Imt R9 te 1 1 8 Image File Formats unes exe oS 119 Multidimensional data 1 1 10 Command Finde aee e ep Peas 1 111 Visualization of Multidimensional Images 1 112 Resampling images Shrinking and Enlarging 1 1 13 ASSIGNMENTS vo ete a re eee ER CGR TACENS Mosa ee ees Se ete de b EM eal a 12 Histogram s 4 Boe OE on Oeo exe s 1 2 2 Region of Interest ROI iate eem eem 1 2 3 Intensity Measurement tee er Be 1 2 4 Image transformation Enhancing Contrast 10 14 17 18 24 27 29 40 41 47 EMBL CMCI Image Basic Course CONTENTS 1 3 1 4 1 5 1 2 5 Image correlation between two images co localization Plora uos enc ee be te eS ale tk gx des 61 1 2 6 ASSIGNMENTS Meuse e acs ue es 62 Filtering So ei 9o 9 Bans icu SOR Bp ERREUR S 63 LOL Convolution s uox wie eee XS RR s 63 L32 Kernels du Wee 3e qeu EE ub exa 67 1 3 Morphological Image Processing 72 1 3 4 Morphological processing Opening and Closing 76 1 3 5 Morphological Image Processing Gray Scale Images 77 1 3 6 Background Sub
53. 200 size above 200 pixel area will be excluded from the measurement Circularity use default 0 1 0 e Show Outline Check box Check Display Results to display the result table Check Clear Results to refresh the measurement record ings Check Exclude on Edges to exclude particles touching the edge of the image Then Click OK Two windows pop up One is the outline image of the measured particles with number labels These numbers cor respond to the numbers you see at the left most column in the result window which also popped up Examine the outline image At the bottom there seems to be some particles which are counted but only part of the particles are included 201t is also possible to do the measurement without redirecting for example you could simply image threshold and while the image is highlighted with red do not click Apply button and do Analyze particle This will then detect particles which are highlighted 107 EMBL CMCI Image Basic Course 1 4 Segmentation 252x252 pixels 8 bit 62K E sus aera EN Circularity 0 00 1 00 Show Outines T Display Results v Exclude on Edges fv Clear Results T include Holes Summarize IF Record Starts T Add to Manager Lok cancer 6 oe oa As hera flus Figure 1 82 Multiple particle Analysis QUESTION How can we eliminate edge touching particles from the analysis Which parameter in the Analyze Par
54. 6 pixels Try other values for the radius parameter Go back to these parameters radius 3 cutoff 0 percentile 0 1 default click on preview detected It is obvious that there are more real particles in the image that were not detected Notice that the detected particles are much brighter then the ones not detected Since the score cut off is set to zero we can rightfully assume that increasing the percentile of particle intensity taken will make the algorithm detect more particles with lower intensity The higher the number in the percentile field the more particles will be detected Try setting the percentile value to 2 After clicking the preview button you will see that much more particles are detected in fact too many particles you will need to find the right balance In this movie percentile value of 0 6 will give nice results Remember There is no right and wrong here it is possible that the original percentile 0 1 will be more suitable even with this film if for example only very high intensity particles are of interest Set the parameters to radius 3 cutoff 1 percentile 0 6 click on preview detected Change the cutoff parameters back to its default 3 and click preview again Notice the particles marked in the two pictures 1 144 tutorialO000 228x275 pixels 8 bit 8 With cutoff 3 both particle are discriminated as non particles and when cutoff 1 only one gets discriminated
55. 83 4 2109 17 Oct 2002 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 12324428 22235482 0006 3495 Journal Article K Miura Tracking movement in cell biology In J Rietdorf editor Advances in Biochemical Engineering Biotechnology volume 95 page 267 Springer Verlag Heidelberg 2005 Kota Miura and Jens Rietdorf Cell Imaging chapter 3 Image quantification and analysis pages 49 72 Scion Publishing Ltd 2006 K Murase T Fujiwara Y Umemura K Suzuki R Iino H Yamashita M Saito H Murakoshi K Ritchie and A Kusumi Ultrafine mem brane compartments for molecular diffusion as revealed by single molecule techniques Biophys J 86 6 4075 93 Jun 2004 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 15189902 0006 3495 Journal Article R J Ober S Ram and E S Ward Localization accuracy in single molecule microscopy Biophys J 86 2 1185 200 Feb 2004 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 14747353 0006 3495 Evaluation Studies Journal Article J B Pawley Handbook of biological confocal microscopy volume 13 of Language of science Springer 2006 ISBN 9780387259215 doi 10 1117 1 600871 URL nttp link aip org link JBOPFO 13 029902 1 135 EMBL CMCI Image Basic Course REFERENCES
56. Kota Miura EMBL CMCI course I ver 2 1 1 Centre for Molecular amp Cellular Imaging EMBL Heidelberg Abstract Aim students acquire basic knowledge and techniques for handling dig ital image data by interactively using ImageJ NOTE this textbook was written using the Fiji distribution of ImageJ IJ ver 1 47n JRE 1 6 0_45 Exercises are recommended to be done using Fiji since most of plugins used in exercises are pre installed in Fiji Many texts images and exercises especially for chapter 1 3 and 1 4 were converted from the textbook of Matlab Image Processing Course in MBL Woods Hall which are originally from a book Digital Image Processing using Matlab Gonzalez et al Prentice Hall I thank Ivan Yudushkin for providing me with the manual he used there Deconvolution tutorial was written by Alessandra Griffa and appears in this textbook with her kind acceptance to do so Figures on stack editing are drawn by S bastien Toshi for his course and appear in this textbook for his kind offer I am pretty much thankful to his figure and am impressed with the way he summarized all the commands related to this S bastien Toshi also reviewed the article in detail and made may suggestions to improve the text I thank him a lot for his effort This text is progressively edited Please ask me when you want to dis tribute Compiled on November 21 2013 Copyright 2006 2013 Kota Miura http cmci embl de Contents 1 1 1
57. Koumoutsakos 2005 There is also a detailed tutorial available in the web site Particle Tracker Tutorials also added to this textbook in the Appendix 1 6 10 so refer to it as well for the following exercise Set correct dimensions of the image by Image Properties Image stacks are by default taken as a z series and not t series Set Slices to 1 and Frames to appropriate size number of frames Start the ParticleTracker plugin Start the particleTracker by Plugins gt Mosaic gt ParticleTracker 2D 3D Study Dot Detection Parameter 127 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series This tracking tool has two parts First all dots in each frame are de tected and then dots in successive frames are linked The first task then is to determine three parameters for dot detection There are three parameters Radius Expected diameter of dot to be detected in pixels CutOff Cutoff level for the none particle discrimination criteria a value for each dot that is based on intensity moment order 0 and 2 Percentile Larger the value more particles with dark intensity will be de tected Corresponds to the area proportion below intensity his togram in the upper part of the histogram Try setting different numbers for these parameters and click Preview Detected Red circles appear in the image stack You could change the frames using the slider below the button After some trials set
58. a swimming or cell 129 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series movement within tissue Berg 1993 Saxton and Jacobson 1997 Kusumi et al 1993 Witt et al 2005 Suh et al 2005 The most basic equation that describes the diffusion is lt r gt 4DT 1 21 where 7 is the mean square displacement D is the diffusion coefficient T is the time scale The MSD is calculated by squaring the net distance a molecule moved during the period of time t For example if an object moved from 0 0 to x y during a period of t the square displacement will be r x y 1 22 r 4 x2 42 V4DT 1 23 If we have many samples then we can average them and the mean square displacement MSD will be or lt r gt lt x y gt 1 24 The equation 1 24 tells us that the MSD 7 is proportional to t which means that when MSD is plotted against various T the plot will be a straight line with a slope 4D Fig 1 92 plot a If the mobility of the target object is lower then the slope becomes less steep Fig 1 92 plot d The mobility of molecules is not always a pure diffusion In the case when there is constant background flow such as laminar flow in the medium molecules drift in a certain direction This is called diffusion with drifts also called biased diffusion Curve b in the Fig 1 92 is a typical MSD curve of this type Since the flow causes the movement vt where v is the 24This equati
59. above 255 are now considered as saturated which means that what ever the value is numbers larger than 255 becomes 255 mshi o 320x510 pixels 8 bit 159K Figure 1 7 m51 image converted to 8 bit without scaling 13 EMBL CMCI Image Basic Course 1 1 Basics of Basics o Distance pixels 487 o ust Save copy x 250 v 255 List save copy a b Figure 1 8 a Intensity profile of 1 3b If the conversion was done with scaling then the profile would look like b When you perform a conversion very different results could appear de pending on how you scale like we have just seen But in many cases you do not recognize such changes just by looking at the image for this rea son one should keep in mind that the conversion may saturate or cause artifacts in the image screwing up scientific images to non scientific ones 1 1 4 Math functions A digital image is a matrix of numbers We can calculate images like usual math If there is a flat image with pixel value of 10 and if you add 1 to the image then all pixel values become 11 We think about a pixel at 5 10 and we write down the calculation as follows f 5 10 10 1 1 2 5 10 f 5 10 1 11 1 2 We generalize this x and y are the coordinates within the image g x y foo y 1 1 3 The original image is f x y and the result after the addition is g x y
60. ading Perim With IJ1 44f and later the perimeter of a composite selection is calculated by decomposing it into individual selections Note that the com posite perimeter and the sum of the individual perimeters may be different due to use of different calculation methods Bounding Rectangle The smallest rectangle enclosing the selection Uses the headings BX BY Width and Height where BX and BY are the coordinates of the upper left corner of the rectangle Fit Ellipse Fits an ellipse to the selection Uses the headings Major Minor and Angle Major and Minor are the primary and secondary axis of the best fitting el lipse Angle is the angle between the primary axis and a line parallel to the X axis of the image The coordinates of the center of the ellipse are dis played as X and Y if Centroid is checked Note that ImageJ cannot calculate the major and minor axis lengths if Pixel Aspect Ratio in the Analyze gt Set Scale dialog is not 1 0 There are several ways to view the fitted ellipse 1 The Edit gt Selection gt Fit Ellipse command replaces an area selection with the best fit ellipse 2 The DrawEllipse macro draws destructively the best fit ellipse and the major and minor axis 3 Select Ellipses from the Show drop down menu in the particle ana lyzer Analyze gt Analyze Particles and it will draw the ellipse for each particle in a separate window Shape Descriptors Calculates and displays the following s
61. ager Activate all ROI by clicking the ROI name in the list while pushing down the SHIFT key Then click Measure button All five ROI will be measured at once Average the values and describe the results in text Assignment 1 2 3 1 Optimize the contrast of image m51_8bit tif Be careful not to satu rate the pixels so you don t throw away important information in the lower pixel values Check the histogram Don t close the histogram window 2 After applying the adjustment above clicking Apply button check the histogram again Compare the histogram before and after What happened Discuss the reason 62 EMBL CMCI Image Basic Course 1 3 Filtering 13 Filtering Filtering can improve the appearance of digital images It can help identi fying shapes by reducing the noise and enhancing the structural edge Not only for the appearance filtering improve the efficiency of segmentation which we will study in the next section Segmented image could be used as a mask to specify regions to be measured Note that the filtering alters the image so one must realize that in most cases processed images cannot be used for quantitative intensity measurement without precise knowledge of how the processing affects the numerical values There are two different types of filtering one involves the frequency do main of images Fourier transformed image while the others deals with spatial domain We first study the spatial filtering
62. alue in the two dimensional image at the same position Max Intensity Min Intensity e Sum of Slices e Standard Deviation Median Question 1 Which is the best method for knowing the shape of the chromosome Question 2 Discuss the difference of projection types and why there is such difference Max Intensity projection is the most frequently used projection method in fluorescence microscopy images as this method picks up bright signals 44 EMBL CMCI Image Basic Course 1 1 Basics of Basics sum of slices method returns a 32bit image since results of addition could possibly exceed 8 bit or 16 bit range Note 1 If your data is a 3D time series hyper stack there will be a small check box in the projection dialog All Time Points If you check the box projection will be done for each time point and the result will be a 2D time series of projections Note 2 The projection of a 2D time series can also be performed Thought the command name is Z Projection the principle of projection is identical regardless of the projection axis Orthogonal Views The projection is a convenient method for visualizing 3D stack on 2D plane but if you want to preserve the original dimensions and still want to visu alize data one way is to use a classic 2D animation This could be achieved easily with the scroll bar at the bottom of the each stack But this has limita tion we could only move in the third dimension
63. am eters are important Mean Gray Value Average gray value within the selection This is the sum of the gray values of all the pixels in the selection divided by the number of pixels Reported in calibrated units e g optical den sity if Analyze Calibrate was used to calibrate the image For RGB images the mean is calculated by converting each pixel to grayscale using the formula gray 0 299red 0 587green 0 114blue or the for mula gray red green blue 3if Unweighted RGB to Grayscale Conversion is checked in Edit Options Conversions Standard Deviation Standard deviation of the gray values e Min amp Max Gray Level Minimum and maximum gray values within the selection Integrated Density The sum of the values of the pixels in the image or selection This is equivalent to the product of Area and Mean Gray Value The Dot Blot Analysis example demonstrates how to use this measurement to analyze a dot blot assay 56 EMBL CMCI Image Basic Course 1 2 Intensity A short note on image based fluorometry In biochemical experiments sci entists measure protein concentration by measuring absorbance of UV light or by labeling proteins with dyes or fluorescence and measure the intensity of emission This is because light absorbance or the intensity of fluores cence intensity is proportional to the density of protein in the irradiated volume within cuvette Very similar to this when fluorescence image of cells
64. and its central concept convolution and walk through various types of convolutions We then study the frequency domain filtering FFT In the FFT world convolution of an image with filter kernel could be done simply by multiplication be tween two images 1 3 1 Convolution In the Process menu we see a long list of filters Most of them are linear filters and can be implemented as convolution as we will shortly see To perform a convolution a small mask called kernel is moved over the image pixel by pixel and apply operations involving the surrounding pixel values see Fig 1 43 The result of operation is written to that pixel position in the output image To understand how the convolution is done we take a one dimensional example Fig 1 44 We have a small 1D array f and we want to convolve a kernel w i We first rotate the kernel by 180 degrees that the order is now reversed Then we align f and w to match the position of the last element of kernel to the first element of f Since we want to have all the kernel elements to have corresponding partner we pad f by 0 This is just for the convenience of 63 EMBL CMCI Image Basic Course 1 3 Filtering Origin Image f x y x Figure 1 43 An image and a Kernel figure taken from DIP Correlation Convolution Origin f w 2 Origin f w rotated 180 00010000 12 32 0 00010000 0 273 2 T i b 00010000 00010000 j 1 2 3 2 0 0 23 2 1 t Starting
65. annel by simply selecting a channel with the horizontal scroll bar and do the process ing directly such as drawing square ROI and deleting that part in that channel in composite view 11 7 Look Up Table We now look at how the matrix of numbers is converted to an image Let s think about a row of pixels with increasing pixel values from 0 to 255 so there are 256 pixels in this row Computer monitor will show a gradient of intensity that is linearly increasing its brightness from black to white This is because the software is giving a command to the monitor such that this pixel x y is 158 so the corresponding voltage required for this position x y in the screen should be mV For this command to be composed software needs a so called look up table LUT The default LUT is the gray scale that assigns black to white from 0 to 255 in the 8 bit image In the 16 bit image gray scale will be valued from 0 to 24 EMBL CMCI Image Basic Course 1 1 Basics of Basics 65535 between black and white LUTs are not limited to such grayscales it could be also colored We call such colored LUT as pseudo color In case of the spectrum LUT GRRE and OSEE see 1 21 The most important point to understand the LUT concept is that with same data the appearance of the image changes when different LUT is used Look Up Table Gray Scale 0 0 0 0 100 0 0 0 0 Look Up Table Sp
66. are recorded in the first column of the results table e g mri stack tif 9 For renamed se lections Edit gt Selection gt Properties y or selections measured via ROI Manager s measure command see ROI Manager the selection label is appended e g blobs gif 0339 0163 or blobs gif mySelection Invert Y Coordinates If checked the XY origin is assumed to be the lower left corner of the image window instead of the upper left corner see also Image gt Properties P Scientific Notation If checked measurements are displayed in scientific notation e g 1 48E2 Redirect To The image selected from this pop up menu will be used as the target for statistical calculations done by Analyze gt Measure m and Analyze gt Analyze Particles commands This feature allows you to outline a struc ture on one image and measure the intensity of the corresponding region in another image Decimal Places 145 EMBL CMCI Image Basic Course 1 6 Appendices This is the number of digits to the right of the decimal point in real numbers displayed in the Results table and in Histogram windows 146 EMBL CMCI Image Basic Course 1 6 Appendices 1 6 5 App 3 Edge Detection Principle Quote from MBL manual One way to find boundaries of objects is to detect discontinu ities in intensity values at the edge of a region These disconti nuities can be found by calculating the first and or second or der derivatives
67. are taken pixel values which is the fluorescence intensity are pro portional to the density of the labeled protein at each pixel positions For this reason measurement of fluorescence intensity using digital imaging could be considered as the two dimensional version of the conventional fluorometry Exercise 1 2 3 1 Open a sample image cells_Actin tif Before actually executing the measurement do Analyze gt Set Measurements A dialog win dow opens There are many parameters listed in the upper half of the window The parameters we select now are Area Integrated Density e Mean Gray Value Check the box of these three parameters Integrated density is the sum of all the pixel values and Mean Gray Value is the average of all the pixel values within ROI So IntegratedDensity Area x MeanGrayValue Select one of the cells in cell_Actin tif image and zoom it up using magnifying tool Switch the tool to Polygon tool Draw polygon ROI around the cell Then do Analyze gt Measure A window titled Results pops up listing the measured values Check that the integrated density is the multiplication of area and the mean gray value This value is not the actual intensity of the fluorescence within the cell since it also includes the background intensity offset Measure the 57 EMBL CMCI Image Basic Course 1 2 Intensity Set Measurements x V Area MV Mean Gray Value Standard Deviation J Modal Gray
68. ated by visual inspections But with digital im ages it is possible to evaluate the degree of co localization more quanti tatively This is done by plotting a so called co localization plot To do this in ImageJ one could download a plugin Colocalization Finder and install it The level of colocalization could be then parametrized by using statistical values such as Pearson s coefficient and Mander s coefficient These val ues have advantages and disadvantages depending on the image proper ties For detailed description on these issues refer to Bolte and Cordeli res 2006 Additional insights pitfalls and tips on localizing spot signals could be found in Waters 2009 This paper also provides detailed examination of the precision of dot detections H download from httpi rsb info nih gov ij plugi amp ss colocalization finder html 61 EMBL CMCI Image Basic Course 1 2 Intensity 1 2 66 ASSIGNMENTS Assignment 1 2 1 Suppose that you have an 8 bit grayscale image showing three objects of distinct intensities against a bright background What would the corre sponding histogram look like Assignment 1 2 2 With image cell_Actin tif do the measurement with 4 cells and one back ground in image as we did in the exercise This time store ROI in ROI manager refer to 1 2 2 Roi and do the measurement at once First you store 5 different ROI one by one Add button Then click Show all but ton in the ROI man
69. chers here and add some description about each of them For more details please refer to an article by Walter et al 2010 MatLab Mathworks With Matlab you could access images as numerical matrix to do process ing and analysis of images Programming is possible with Matlab scripting language Scripts could be kept as files and execute them directly from the Matlab command line These files are called m files Many imaging re lated tools are publicly available via Internet download Scripts could be exported as execution files and could be distributed without Matlab itself as these stand alone execution files only require freely available Matlab li brary A free alternative is Octave I have never tried this yet but this freeware is under extensive development and worth for some trial Imaris Bitplane Imaris is also a commercial software especially powerful in interactive 3D visualization 3D time course sometimes called 4D with multiple chan nels then this would be called 5D could also be visualized and interac tively adjusted with their appearance Some analysis packages are option ally available I sometimes use optional package Imaris Track for track ing spotty signals Algorithm for linking particles is excellent uses graph theory based algorithm developed in Danuser lab In EMBL you could try using this software by so called floating license which enables you to use the software from any computer wit
70. d others you can change the magnification factor Before clicking the Focus on Selected Trajectory button go to View Preferences menu in the results window and select the Magnification Factor option Select magnification of 2 4 Click on the Focus on Selected Trajectory button to see the size of the new window Close the window Fourth Step re linking the particles To explain the re linking option we will use a different data sample Artificial zip Close all windows including the original movie window Load the new image sequence from Artificial zip and start the plugin Set the particle detection parameters to radius 3 cutoff 3 0 percentile 0 9 Set the particle linking parameters to link range 1 displacement 10 Start the algorithm and when it s done click the View all Trajectories button Zoom in on the overview window for better view Select an area of interest to include the 2 adjacent blue and grey trajectories as shown here ny rajectories Visual 40096 1 100 test linear 03 001 118x11 8 pixels 16 bit grayscale 2 7 MB enna rms E70 EE l Filter Options Increase the magnification factor to 8 and focus on that area Scroll through the focused view or view it with animation Image gt Stacks gt Start Animation It seems that these two trajectories are actually 1 longer trajectory Why was it recognized as 2 trajectories Scroll to slice frame 71 At this point the de
71. ded to 0 and 0 56 to 1 Thus the bit depth limits the resolution of the analog to digital conver sion AD conversion Higher bit depth generally allows higher resolution ImageJ has a high bit depth format called signed 32 bit floating point im age In all above cases with 8 bit and 16 bit the pixel value is represented in integer but floating point type enables decimal points real number for the pixel value such as 15 445 Though 32 bit floating point image can be used for image calculation many functions in ImageJ do not handle them properly so cares should be taken when using this image format If you want to know more about the 32 bit format read the following box a bit complicated you could just pass through 32 bit FLOATING POINT images utilizes efficient use of the 32 bits Instead of using 32 bits to describe 4 294 967 296 integer numbers 23 bits are allocated to a fraction 8 bits to an expo nent and 1 bit to a sign as follows V 1 S 1 F x 2 E 127 whereby S Sign uses 1 bit and can have 2 possible values F Fraction uses 23 bits and can have 8 388 608 possible values E Exponent uses 8 bits and can have 256 possible values Practically speaking this allows for an almost infinite number of tones between level 0 and 1 more than 8 million tones between level 1 and 2 and 128 tones between level 65 534 and 65 535 much more in line with our human vision than a 32 bit integer i
72. dimensional data takes a form of multiple single image frames with numbering such as image0001 tif image0002 tif image0003 tif We can import such numbered files recursively and create a stack in ImageJ Exercise 1 1 9 1 We import multiple image files as an image stack Download a zipped fileby EMBL gt Samples gt Spindle Frames zip Unzip the file by double clicking Then File gt Import gt Image Sequence will open a dialog window and you must specify the first file of the image series Select sample sequence eg5 spindle 50000 tif Then another dialog window pops up Fig 1 24 6We do also have the stack directly downloadable but we try to load numberd tif images here 30 EMBL CMCI Image Basic Course 1 1 Basics of Basics _ Sequence Options Number of images Starting image Increment Scale images File name contains or enter pattern Convert to 8 bit Grayscale C Convert to RGB V Sort names numerically Use virtual stack 512 x 512 x 5 5 0MB Help Cancel OK Figure 1 24 Importing multiple frames as a Stack Image automatically detects the number of files in the folder and then another window opens to ask for the number of images you want to import the starting image the increment between the num bering of the files and also the common part of the names of the files you want to import For example if your sequence starts with ex
73. e Basic Course 1 1 Basics of Basics a Original stack first frame This stack b Color coded stack consists of 72 frames frame c Scale of the time color coding From frame 1 to 72 corresponding color is shown as a scale Figure 1 33 Temporal Color Coding In this way we can pick up a column of voxels from every XY positions compute statistics and create a two dimensional image with each of its XY position filled with voxel column statistics This two dimensional image is the result of the projection along Z axis or what we call Z projection Projection could also be in other axes not only along Z axis If we do a projection along X axis we would then have a projection of the data to YZ plane Projecting aloing Y axis will result in a projection to XZ plane this relates to the orthogonal viewing which we try in the next section Exercise 1 1 11 2 Open image mitosis_anaphase_3D tif Then do all types of projec tions you could choose by Image gt Stack gt Z projection There are Average Intesnity 43 EMBL CMCI Image Basic Course 1 1 Basics of Basics i i ANE EN i TIJST LJ Eq coq fj J fj jf p Re VVVVV VV VV VY Figure 1 34 A three dimensional stack can be regarded as a gridded cube The projec tion calculates various sorts of statistics for each two dimensional positions which can be viewed as a stack of voxels and store that v
74. e expected size of the image file in bytes 1 byte 8 bit Create a new image with the dimension as above and save the image in bitmap bmp format and check the file size Is it same as you expected or different Save the same image in text file format and check the file size again Assignment 1 1 7 Resizing 1 Enlarge the sample image 4pixelimage sample tif by 150 while the Interpolation check box in the size adjustment window is ON Study the pixel values before and after the enlargement What happened Describe the result 2 Change canvas size by Image gt Adjust gt Canvas Size forany image What s the difference to Resize 51 EMBL CMCI Image Basic Course 1 2 Intensity 12 Intensity An image has only one type of information a distribution of intensity The image analysis in biology deals with this distribution in quantitative ways We investigate the distribution from different angles using various algo rithms and analyze biological phenomena such as shapes cell movements and protein protein interactions For example in GFP labeled cells inten sity of signal is directly related to the density of the labeled biological com ponent We now start studying how to interpret signals and how to extract biologically meaningful numerical values out of intensity distribution 12 1 Histogram If there is an 8 bit image with 100 pixel width and 100pixel height there are 10 000 pixels Each of these pixels has
75. e image we use in this exercise z should be 8 and frames should be 46 39 EMBL CMCI Image Basic Course 1 1 Basics of Basics 8 OO mitosis tif 150 c 1 2 z 3 5 t 1 51 15 13x17 35 um 171 Z b lt Figure 1 31 A Hyperstack data showing spindle dynamics This 5D data could be down loaded by File gt Open Samples gt Mitosis series at a same constant z position or go through z slices at a certain time point To go back to the normal stack mode use Image gt Hyperstacks gt Hyperstack to Stack 1 1 10 Command Finder Commands for these stack related tools especially editing tools reside deep inside the menu tree and it is not really convenient to access those commands by manually following the menu tree using mouse For such a case and if you could remember the name of the command a quick way to run the command is to use command finder 40 EMBL CMCI Image Basic Course 1 1 Basics of Basics There is a short cut key to start up the command finder control l in win dows and in mac OSX command l On start up the command finder inter face looks like figure 1 32a Type in the command that you want to use in the text field labeled search and then menu item is filters as you type in fig 1 32b Select the one you want and then click the button run This is the same as selecting that command from the menu tree This tool is also useful when you know the name of a command but for
76. e this to csv CSV stands for comma separated file and this is more classic but general data format which you could easily import in many software such as R 1 5 8 Summarizing the Tracking data Tracking of a moving object results in a list of position coordinates This list can be saved as a text file and imported to spreadsheet software such as Excel The instantaneous velocity is derived by calculating the distance between consecutive time points For example if the position of an object at time point f is x y and the object moves to a position X 41 Y 41 in the next time point t At then the instantaneous velocity v is a V xia i yii yi 1 20 For each frame the instantaneous velocity can be calculated One way to summarize the data is to average the resulting instantaneous velocities and append its standard deviation In most cases in biology velocity changes with time and this dynamics can be studied by plotting instantaneous ve locity vs time on a graph Such plotting reveals characters of the move ment such as acceleration kinetics or periodicity Movements within or ganisms can be both random and directed To make a clear distinction between these different types of movements mean square displacement MSD plotting is a powerful method Although this method has been ex tensively used in studying molecular diffusion within the plasma mem brane it can also be used in other scales such as bacteri
77. e x0 y0 144 e z0 192 e Ar 89 nm e Az 290 nm e Nx 288 e Ny 384 e Nz 32 e Perform a visual verification of the Optic PSF stack after that you can discard it Note that the voxel size of the acquired image is coherent with the resolution of an objective with 1 4 NA for the used wavelength Open the Deconvolution plugin In the PSF section select PSF for deconvolution In the Algorithm section select the Richardson Lucy algorithm with 20 iterations Compute maximum intensity projections of the original and deconvolved stacks Compare them If there is some time left try to perform a deconvolution with another algorithm e g the Tikhonov Miller algorithm and comment the results You may want to activate Test the algorithm on a small rectangular selection You could also study the influence of using a PSF with wrong parameters for example change the NA value or introduce spherical aberrations
78. each channel Uncheck Create Composite and check keep source images Then try swap ping color assignments to see the effect c Working on each channel separately Close all windows and open the RGB cell tif again Do Image Color Channel Tools Then click button More and select Create Composite 19 EMBL CMCI Image Basic Course 1 1 Basics of Basics Eoix 162x181 pixels RGB 115K Composite v Channel 1 v Channel 2 v Channel 3 More Figure 1 12 Channel Tool Resulting image is a three layer stack and each layer corresponds to one of R G or B Each layer can be processed individually RGB cell tif 25 Bl x 1 3 Red 1344x1024 pixels 8 hit 3 9MB Figure 1 13 Composite image Note slider at the bottom for switching between three channels Using Channel Tools again 20 EMBL CMCI Image Basic Course 1 1 Basics of Basics Perr l Channel 1 v Channel 2 l Channel 3 More Figure 1 14 Channel Tool now only selected for Channel 2 Choose color from the pull down tab instead of Composite Select channel 2 in this image this will be Green channel Select a part of the image using a rectangular ROI RGB cell tif 25 2 3 Green 1344x1024 pixels 8 bit 3 3MB Figure 1 15 ROI selection in channel 2 Note the position of slider Then do Edit Clear This will pop up a window 21 EMBL CMCI Ima
79. ectrum Figure 1 21 Look Up Table is a table that defines appearance of pixel values as an image With same pixel values how they are colored could be different which depends on the selection of LUT This is just like a situation when you start checking a menu in a restaurant with limited amount of money in your pocket Say you want to eat a pizza You have only 10 in your pocket Looking at the pizza menu you will not try to find what you want from names of pizza and what are the toppings but instead you will check the prices listed in the right side of the menu trying to figure out which pizza is less then 10 When you find 7 5 in the list then you slide your sight to the left side of the menu and find out that the pizza is Margherita Similar to this software first checks the pixel value and then goes to the look up table menu to find out which bright ness should be passed to the monitor as a command find a convincing price in the menu then sliding your sight to the left and find out which pizza to order Exercise 1 1 7 1 25 EMBL CMCI Image Basic Course 1 1 Basics of Basics For 8 bit images there is a default LUT normally called grayscale To see the LUT open the image Cell_Colony jpg and then do Image gt Color gt Show LUT LUT window pops up showing the relation ship between pixel value and pixel intensity Try to change the LUT by Image gt Lookup Tables gt Spectrum
80. eel S 0l e 0 Outlines Masks Ellipses Original image Thresholded using Image Adjust Threshold Check Display results to have the measurements for each particle displayed in the Results window Check Clear Results to erase any previous mea surement results Check Summarize to display in a separate window the particle count total particle area average particle size and area fraction Check Exclude on Edges to ignore particles touching the edge of the image or selection 163 EMBL CMCI Image Basic Course 1 6 Appendices original image thresholded default settings Show masks Flood Fill Exclude on Edges 393 ar Size 300 99999 pixel 2 Size 1 300 pixel 2 Size 200 99999 Circularity 0 8 1 0 l Circularity 0 6 Check Flood Fill and ImageJ will define the extent of each particle by flood filling instead of by tracing the edge of the particle using the equivalent of the wand tool Use this option to exclude interior holes and to measure particles enclosed by other particles The following example image con tains particles with holes and particles inside of other particles 164 EMBL CMCI Image Basic Course 1 6 Appendices 1 6 8 App 6 Image Processing and Analysis software scripting language Image is not only the tool for scientific image processing and analysis I list some other software and tools used by EMBL resear
81. eft corner image Expression parser va2 0 inis Expression Enter sn expression using canonical mathematical a functions and capital single letters ss variable specifying the chosen image ImgLib algorithms are also supported Examples At microtubule tif v aA a A B 30 a sqrt A 2 B 2 cos C B microtubule 1 tif w Figure 1 76 Image Expression Parser Fiji To compare the original and the edge detected images go back to the image microtubule and do File Revert Then with Result 102 EMBL CMCI Image Basic Course 1 4 Segmentation of Microtubule do Image gt Type gt 8 bit to scale down the bit depth Merge the images by Image gt Color gt RGB Merge In the following dialog window choose appropriate color for each im age such as shown below x Red Resultof microtubule Green None A Blue microtubule tif h2 b Figure 1 77 RGB merge of detected edge over the original image a Dialog for assign ing channels b Example of merged color image with original image in blue and edge detected image in red 1 4 3 Morphological Watershed In the previous sections we discussed segmentation based on threshold ing and edge detection Morphological watersheds provide another way to segment objects One advantage of the watershed algorithm is that it out puts continuous segmentation boundaries and often tends to produce sta ble segmentation results In ImageJ watershed
82. egmentation 1 4 6 ASSIGNMENTS Assignment 1 4 1 Quantum dots Devise a segmentation strategy to segment the following image of quantum dots quantumdots tif Extract the following parameters from the image 1 number of dots 2 histogram of areas 3 histogram of integrated densities You can draw histogram of the results in the Results window by using a Edit gt Distribution function associated with the Result window e s Assignment 1 4 2 DIC images of cells Find the boundary of this cell dic_cell tif using any of your favorite image segmentation techniques 112 EMBL CMCI Image Basic Course 1 4 Segmentation Assignment 1 4 3 Actin filaments Devise an image processing strategy to obtain the distribution of filaments from this image actin tif and subsequently calculate 1 the mean filament length 2 the variance in filament length 3 the number of filaments Note image segmentation may be complicated by the low light levels in the image Assignment 1 4 4 Fixed cells 113 EMBL CMCI Image Basic Course 1 4 Segmentation Here are some cells that are fixed and stained for actin red tubulin green DNA blue and a histone marker not shown 4color_cells Devise an image processing strategy to segment the cells You may operate on any of the color channels in the image or a multiple of them This problem is especially tricky because many of the cells are touching
83. ency signals such as noise will be mapped further from the origin Typical noise has no spatial bias so the noise signal will appear all over the power spectrum FFT image Anisotropic patterns in original image will results in anisotropic signal in 2D power spectrum e g horizontal pattern will cause horizontally aligned high intensity peaks 1 3 10 Frequency domain Convolution Convolution we studied in section 1 3 1 is a sequential procedure The con volution kernel is applied every time you slide the position of the kernel by one pixel in x or y direction This processing is much simpler when per formed in the FFT domain it can then implemented as a multiplication We experience this in the exercise below Exercise 1 3 10 1 Convolution in Frequency domain Edit Laplacian filter kernel using the convolution interface Process 87 EMBL CMCI Image Basic Course 1 3 Filtering gt Filter gt COnvolution 3x3 Laplacian kernel looks like this 0 1 0 1 4 1 0 1 0 Save the kernel somewhere in your computer as laplace3x3 txti Open sample image blobs tif by File Open and then convert blobs gif IN ini xi 256x254 pixels 8 bit inverting LUT 64K ww Figure 1 62 blobs gif the image to 32 bit Image gt Type gt 32 bit then increase the canvas size to 256 x 256 Position should be cen n ter Image gt Adjust gt Canvas Size Duplicate the image so that we
84. ercise 1 3 3 1 Load noisy fingerprint tif and broken text tif Apply dilation or ero sion several times by setting the number of iterations To change this number use Process gt Binary gt Options Bi nary option window opens and you can set several parameters Count matters with setting the number of overlapping pixels between struc turing element and the object that determines output 0 or 1 Larger count causes less degree of erosion or dilation From ImageJ manual Iterations specifies the number of times erosion dilation opening and closing are performed Count specifies the number of adjacent background pixels necessary before a pixel is removed from the edge of an object during ero sion and the number of adjacent foreground pixels necessary before a pixel is added to the edge of an object during dilation Check Black Background if the image has white objects on a black background 73 EMBL CMCI Image Basic Course 1 3 Filtering 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 The structuring element translated to these locations does not overlap any 1 valued pixels in the original image 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
85. ernel using the surface plot Analyze gt Surface Plots Assignment 1 3 2 Morphological Image Processing Load example image EMBL gt Smaple Images NPC TOI tif Split chan nels and work on nucleus channel in te following red channel Design an image processing protocol to create a binary image of nucleus boundary Write down the protocol as a flow chart and also show the resulting image some important tips use Process gt Binary gt Convert to Mask use erosion dilation and image calculator 94 EMBL CMCI Image Basic Course 1 4 Segmentation 14 Segmentation Segmentation refers to the process in which an image is subdivided into constituent regions or objects These objects can be further processed or analyzed for the extraction of quantitative information Biological image data is usually messy and noisy and as a result difficult to segment prop erly Multiple image pre processing steps are often required to allow a good segmentation We often combine segmentation with various morphologi cal processing and filtering techniques such as the ones described in the previous section to achieve an accurate and robust segmentation of an im age Image segmentation algorithms are generally based on one of two ba sic properties of intensity values discontinuity and similarity In the first category the approach is to partition an image based on abrupt changes in intensity edge detection algorithms falls i
86. ertical like shown in the figure below Then change the brightness this will change the x position of the vertical LUT and at the same time you will see the ratio of black and white area changes This corresponds to the changing of thresh olding value PEE lolx Minimum RUPEE aa el Maximum 4 gt Brightness C lH Contrast Auto Reset Figure 1 72 Thresholding by using Brightness Contrast Control Dialog Threshold could be also done by setting two threshold values lower limit and upper limit which means that pixels with a certain range of values Some of Process operations work only with binary images so thresholding is a prerequisite for those filters 96 EMBL CMCI Image Basic Course 1 4 Segmentation could be selected This operation is sometimes called density slice We then have a new rule as follows with lower threshold value T and upper threshold value T g x y l rI 1 9 0 otherwise Exercise 1 4 1 2 Open the image 2D Gel jpg or revert the image to the original by File gt Revert if you still have the image used in the precious ex ercise Then do Image gt Adjust gt Threshold You will see that the Gel image is automatically thresholded The area highlighted in red is where pixels with value lower larger than 0 and lower than 149 In the Threshold window the range of values that is high lighted is shown by red rectangular frame over the histogram Try changing t
87. es gt TransportOfEndosomalVirus tif Ap ply the color coding to this time lapse movie Image gt HyperStacks gt Temporal Color Code In the dialog choose a color coding ta ble from the drop down list Principle of the coding is same as the look up table and only the difference is that the color assignment is adjusted so that the color range of that table fits to the range of frames fig 1 33 Projection Projection is a way of decreasing an n dimensional image to an n 1 dimensional image For example if we have a three dimensional XYZ stack we could do a projection along Z axis to squash the depth information and represent the data in 2D XY image We lose the depth information but it helps us to see how the data looks like The principle is like this we could think of XYZ data as a cubic entity This cube is gridded and composed of small cubic elements If we have a Z stack with 512 pixels in both XY and with 10 slices in Z we then have a cube composed of 512 x 512 x 10 2 621 440 small cubes These small cubes are called voxels instead of pixels Now if we take a single XY position there are 10 voxels at this XY position figure 1 34 Imagine the column of 10 voxels Each voxel has a given intensity and we can compute statistics over these 10 values such as the mean the minimum the maximum the standard deviation and so on We can then represent this column by a single statistical value 42 EMBL CMCI Imag
88. eview Review Tutorial I F Sbalzarini and P Koumoutsakos Feature point tracking and trajectory analysis for video imaging in cell biology J Struct Biol 151 2 182 195 Aug 2005 doi 10 1016 j jsb 2005 06 002 URL http dx doi org 10 1016 3 7185b 2005 06 002 B J Schnapp J Gelles and M P Sheetz Nanometer scale measurements using video light microscopy Cell Motil Cytoskeleton 10 1 2 47 53 1988 URL http www ncbi nlm nih gov entrez query 136 EMBL CMCI Image Basic Course REFERENCES fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 3141071 89028766 0886 1544 Journal Article G J Sch tz H Schindler and T Schmidt Single molecule microscopy on model membranes reveals anomalous diffusion Biophys J 73 2 1073 80 Aug 1997 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 9251823 97395719 0006 3495 Journal Article Michael Schwarzfischer Carsten Marr Jan Krumsiek and PS Hoppe Efficient fluorescence image normalization for time lapse movies In Microscopic Image Analysis with Applications in Biology Heidel berg Germany September 2 2011 number x pages 3 7 2011 URL http www helmholtz muenchen de fileadmin CMB PDF jkrumsiek Schwarzfischer2011 imageprocessing pdf M P Sheetz S Turney H Qian and E L Elson Nanometre level analysis demonstrates that lipid flow does not
89. fore the comparison takes place xe et seas Quote from http www cs utk edu mclennan anon ftp FCMC tr nodel4 html See the example in Fig 1 45 for convolution with two dimensional matri ces Padded f U 000000t PU UUU 000000 0 C LC Origin of f x y l 0000000 0 0000 00001000 6 00000 w x y 00000000 0100 1 2 3 00000000 0000 456 00000 0 0 00000 78 9 000000 0 0 a b x Initial position for full correlation result same correlation result 1230 00000 00000000 00000 4 56000000 000000 0 9870 7 8 9 000000 000000 6540 00000000 098700 2 2 1 0 0007000 0 06540 0 0 0 0000000 032100 000000000 00000 0 0 000000000 0D0000 0 0000000 0 00000 0 0 fC c d e Rotated w full convolution result same convolution result fo 8 70 00000 00000000 00000 654000000 000000 2370 I3 2 1 000000 000000 45 6 000000000 0 12 50 0 1 7890 00010000 045600 00 00 000000000 078900 000000000 UOU UUU 000000000 00000 0 000000000 000000 f g h Figure 1 45 Two dimensional convolution figure taken from DIP The matrix a is first padded b then starting from the top left corner f the matrix is convoluted g and then the padded rows and columns are removed to return an output matrix with the same dimension as the original h 66 EMBL CMCI Image Basic Course 1 3 Filtering 1 3 2 Kernels In Process menu we have many ope
90. ge Basic Course 1 1 Basics of Basics Process Stack x Process all 3 images There is no Undo if you select Yes No Cancel Figure 1 16 Asking you whether you want to process all channels Click No because you want to process only one channel RGB_cell tif 25 Bl x 2 3 Green 1344x1024 pixels 8 bit 3 3MB Figure 1 17 Channel 2 pixel values inside selected ROI becomes 0 Troubleshooting If the ROI is not cleared becomes bright then you should change the background color setting Do Edit Option Colors and you will see a pop up window like this 22 EMBL CMCI Image Basic Course 1 1 Basics of Basics Foreground white Background black Selection yellow Cancel Figure 1 18 Color selection dialog Make sure that the background is black Do the ROI clear ing again Select Composite in the pull down tab of channel tool BET cix v Channel 1 v Channel 2 v Channel 3 More Figure 1 19 Choosing Composite all channels visual Resulting image should look like below 23 EMBL CMCI Image Basic Course 1 1 Basics of Basics RGB cell tif 25 2 3 Green 1344x1024 pixels 8 bit 3 3MB Figure 1 20 Only channel 2 is devoid of image within the selected ROI In this case the intensity of the green channel in the ROI is now set to 0 clear You could do such processing of single ch
91. ging condition rather than increasing the file size of the image On the other hand if you are try ing to measure protein density a higher bit depth allows you to have more precise measurements so a larger bit depth is recommended A draw back is that it takes longer time for data transferring as the bit depth becomes larger This may in turn limits the time resolution of image sequences The balancing between imaging conditions and the type of analysis afterward is very much coupled and should be thought well before your experiment More details on this discussion could be found in microscopy textbook such as Pawley 2006 The choice of bit depth depends on deals with noise Here is an explanation from S bastien Tosi When minute details need to be preserved in both high and low 3Quantum effeciency is a measure of proportion of photons converted to electric im pluse For example QE of photographic film is ca 10 that of human eye is 20 Recent CCD allows 80 EMBL CMCI Image Basic Course 1 1 Basics of Basics intensity regions of a sample a large bit depth is necessary to ensure a good image quality Indeed for low bit depth if the sig nal level is adjusted to avoid saturation in the brightest regions there is a high risk to end up coding the signal of the faintest re gions over only very few bits which has a very detrimental ef fect on the image quality A large bit depth is hence necessary to provide a fine slici
92. got where it is in the menu tree since it also shows where the command is by showing its menu path a The interface on start up b Typing in some command name will filter the menu items and shows the hits Figure 1 32 Command Finder 11 11 Visualization of Multidimensional Images We have an intrinsic limitation in displaying multidimensional images but we still want to check data by our eyes For this reason ways to visualize multi dimensional data have been developed We go over some of those methods which are frequently used in this section Color Coding With multiple channels we could have several different signal distribu tions per scene from different types of illuminations or from different types of proteins One typical way to view such multiple channel image is to color code each channel e g actin in red and Tubulin in green We have 41 EMBL CMCI Image Basic Course 1 1 Basics of Basics already scene such images in the RGB section 1 1 6 The color coding is not limited to the dimensions in Channels but also for slices Z and time points T By assigning a color that depends on the slice number or time points the depth information within a Z stack or a time point information with in a time lapse movie could be represented as a specific color rather than by the position of that slice or frame within image stack Exercise 1 1 11 1 Open EMBL gt Sampl
93. gure 1 81 A small binary image left and its distance transform right 1 4 4 Particle Analysis After the segmentation the objects could be analyzed for extracting various parameters Powerful function in ImageJ for this purpose is particle analy 105 EMBL CMCI Image Basic Course 1 4 Segmentation sis function It counts the number of objects particles in the image extract morphological parameters measures intensity and so on For details on extractable parameters refer to Appendix 1 6 4 of this textbook Here we take an example to learn the basic use of this function Exercise 1 4 4 1 Single Particle Detection Load Circles tif and segment two circles by watershed operation as we did in the previous section Then select the Wand Tool from the tool bar Click one of the circles Check that a ROI is automatically created at the edge of the circle Automatic detection is done by pixel similarity principle In the above case using the wand tool computer searches for pixels in the surrounding of the clicked pixel for similarity in the pixel value Similar pixels will be labeled Then in the next round computer searches for each of the labeled pixel for similar surrounding This continues until searching hits the boundary Particle analysis uses the same strategy but there will be specific labeling for each particle Exercise 1 4 4 2 Multiple Particle Analysis Open rice tif Before applying particle analysis
94. hape descriptors Circularity Area Perimeter approaches 0 0 it indicates an increasingly elongated shape Values With a value of 1 0 indicating a perfect circle As the value may not be valid for very small particles Uses the heading Circ 143 EMBL CMCI Image Basic Course 1 6 Appendices Aspect Ratio Major Axis Minor Axis If Fit Ellipse is selected the Major and Minor axis are displayed Uses the heading AR The aspect ratio of the particle s fitted ellipse i e e Roundness Area al Major axis Of the inverse of Aspect Ratio Uses the heading Round e Solidity oS Note that the Edit Selection gt Convex Hull command onvex area makes an area selection convex Feret s Diameter The longest distance between any two points along the selection bound ary also known as maximum caliper Uses the heading Feret The angle 0 180 degrees of the Feret s diameter is displayed as FeretAngle as well as the minimum caliper diameter MinFeret The length of the object s projection in the X FeretX and Y FeretY direction is also displayed Integrated Density The sum of the values of the pixels in the image or selection This is equiv alent to the product of Area and Mean Gray Value With IJ1 44c and later Raw integrated density sum of pixel values is displayed under the head ing RawIntDen when Integrated density is enabled The Dot Blot Analysis tutorial demonstrates how to use
95. hat it was the last active image before clicking Run Try to deconvolve the widefield image you simulated using the inverse filtering method In the Algorithm section select Direct inversion and click Run Comment the result 2 2 Noisy inverse and Wiener filtering We will create now an artificially blurred and noisy image more realistic simulation Go back to Convolution in the Algorithm section Activate the Add noise checkbox Play around with both noise models at various noise levels e g between 10 30 and 60 dB to get a feeling for this functionality What are the principal sources of Gaussian and Poisson noise in microscopy images Choose the Gaussian model with a noise level such that you cannot visually distinguish the image from the one obtained at point 2 1 Apply the inverse filtering to restore the image and comment the results Try now the inverse filtering with regularization Wiener filtering on the noisy and blurred image Select Regularized Direct Inversion in the Algorithm section and run the deconvolution with different values of the regularization parameter Lambda How does the regularization parameter influence the result Exercise 3 2D iterative deconvolution Choose the Poisson model with a noise level of 30 dB Select the Richardson Lucy deconvolution algorithm which is one of the simplest deconvolution algorithms for shot noise limited imaging Try out the algorithm with different numbers of iterations n
96. he PSFGenerator plugin in ImageJ Generate a 3D PSF for a widefield microscope with following parameters e ni 1 518 e NA 1 4 e W040 0 e 500nm e amplitude 255 e background 0 e SNR 100 e x0 y0 32 e Z0 64 e Ar Az 50nm e Nx Ny 64 e Nz 128 You can discard the image called PSF for deconvolution Browse through the PSF using the VolumeViewer the 3D Viewer and the Orthogonal Views tools of ImageJ hint change the lookup table of the optic PSF into Fire to better appreciate its shape Change the numerical aperture e g NA 0 8 and or the wavelength e g 400 or 600 nm Generate the corresponding PSFs and describe the influence of these parameters Go back to the initial set of parameters W040 controls the amount of spherical aberrations set its value to 500 nm Describe the effect on the PSF What are possible causes of spherical aberrations Exercise 2 2D inverse filtering Load the images FluorescentCells tif and PSF Defocus tif Launch the Deconvolution plugin 2 1 Noise free inverse filtering We will first create an artificially blurred image using the defocusing blur kernel In the PSF section of the plugin select the image PSFDefocus tif Make sure that Normalize PSF is selected In the Algorithm section select Convolution Make sure that Add noise is deselected Click the Run button note to apply the plugin s functionalities to a specific image you have to make sure t
97. he resulting FFT image See spatial domain images shown in below Stripe frequency decreases from left to right In corresponding FFT images shown in the second row high intensity pixels high lighted in red become closer to the center of FFT image as the frequency of pattern in the original spatial domain image decreases EE oxi 50x50 pixels 8 bit 50x50 pixels 8 bit EES oxi 50x50 pixels EES oxi 50x50 pixels 8 bit c Figure 1 59 Original images Spatial domain images with various frequency l X 64x64 pixels 8 bit a c Figure 1 60 FFT images Frequency domain images of images with various frequency Frequency domain image is a plot with vertical frequency in vertical axis and horizontal frequency in horizontal axis The two axis crosses at the centre of the image A schematic drawing of the FFT domain is shown below to indicate how the FFT image would be distributed depending on the original spatial domain image 86 EMBL CMCI Image Basic Course 1 3 Filtering Figure 1 61 Distribution of Signals in 2D Power Spectrum Frequency of pattern increases from center towards periphery black arrows Direction of pattern is reflected in the align ment of signal in 2D power spectrum Signals with lower frequency such as large objects with smooth transitions will be mapped close to the origin center of the 2D power spectrum im age while higher frequ
98. he upper and lower threshold value using the sliding bar below and study the effects on highlighted area in image Threshold mig o x FT o i Default Red Dark background Auto Any Reset set Figure 1 73 Thresholding Dialog Set button in the threshold window enables to you input the lower and upper threshold numerically The original image file is not al tered until you click the button Apply at the bottom of the threshold window Click Apply and check the result of conversion by saving it 97 EMBL CMCI Image Basic Course 1 4 Segmentation as a text file or simply by checking the pixel values using the value indicator in the status bar Instead of manually setting the threshold level many algorithms for au tomatically setting the threshold level exist Auto button in the threshold window is one of these automatic thresholding algorithms Various algo rithms for automatic determination of threshold value are available such as Otsu Maximum Entropy and so on and you could choose one of them by drop down menu on the left side Following is a list of available Algo rithms IsoData e Maximum Entropy Otsu Mixture Modeling e Huang e Intermodes e Li Mean e MinError e Minimum Moments Percentile e RenyiEntropy e Shanbhag e Triangle e Yen 98 EMBL CMCI Image Basic Course 1 4 Segmentation um Huai Intermodes MaxEntropy Minimum M
99. hin EMBL domain The number of simultaneous usage is limited to three as of June 2010 Another module that gives additional power to Imaris is Imaris XT which enables access ing Imaris objects from Matlab Java or ImageJ I have some example Java codes so if you are interested I could give you as an example Imaris is pretty expensive and apparent disadvantage Python Free 165 EMBL CMCI Image Basic Course 1 6 Appendices Python is not a software package but is a scripting language There are many other scripting languages like Ruby but the merit of Python is that there are numerous libraries for image processing and analysis In terms of scripting accessibility is similar or more powerful than Matlab since its bridging capability to many computer languages such as C C and Java Considering that the trend of image processing and analysis is getting more and more towards cross language library usage Python is a good choice to learn for high end processing and analysis Cell Profiler Free This free software is a bit less with available functions compared to Im age but is easier and robust in constructing pipelines for image process ing and analysis One could interactively construct pipeline and do high throughput processing and analysis for many data 166 EMBL CMCI Image Basic Course 1 6 Appendices 1 6 9 App 7 Macro for Generating Striped images Below is an ImageJ macro code for generating stripes to d
100. ibution should be near flat To get such plot we need to compensate the histogram by shifting values from the bars 5 and 7 to the bars 7 and 9 respectively now bars are in red after shifting After this shifting the cumulative plot now looks more straight and diagonal Fig 1 38b right red curve 53 EMBL CMCI Image Basic Course 1 2 Intensity 9 Original 2170 Equalized count cumulative count pixel value pixel value a b Figure 1 38 a Histogram Equalization Very simple case The actual calculation uses cumulative plots b of the histogram of original image and uses it as a look up table to convert original pixel value applied point by point Exercise 1 2 1 2 Histogram Normalization and Equalization Open sample image g1 tif and then duplicate the image by Image gt Duplicate Click the duplicate and then Image gt Process gt Enhance Contrast A dialog window pops up Set Saturated Pix els to 0 0 and check Normalize while unchecking Equalize His togram then click OK Compare histogram of original and normal ized images Duplicate g1 tif again by Image gt Duplicate Click the duplicate and then Image gt Process gt Enhance Contrast A dialog win dow pops up Set Saturated Pixels to 0 0 and uncheck Normal ize while check Equalize Histogram then click OK Compare his togram of original normalized and equalized images Analyze gt H
101. ic trace already listed in Class listing by double clicking Trainable Segmentation 150 252x252 pixels 8 bit 62K Training Train classifier Toggle overlay Create new class Settings Figure 1 85 After first round of trainable segmentation To create segmented image click Create result A separate win dow with segmented image will open Fig 1 86 During the training a set o feature data and its classes is generated 110 EMBL CMCI Image Basic Course 1 4 Segmentation ioj x 252x252 pixels 8 bit 62K Figure 1 86 Segmentation Result of Trainable Segmentation PlugIn To save this data used for training click save data Saved arff file could be loaded later to segment another image should be rice in this case of course using Load data button OPTIONAL Segmented image could be used as a mask to ana lyze original image To do so open both the original rice tif image and binary image segmented rice tif image Then Analyze Set Measurement and set redirect to drop down menu to original image rice tif Then activate the segmented image window thresh old the image but do not apply and do particle analysis Detection of particles are done in the segmented image and measurements will be redirected to corresponding particle areas in the original image 111 EMBL CMCI Image Basic Course 1 4 S
102. icle M K Cheezum W F Walker and W H Guilford Quan titative comparison of algorithms for tracking single fluo rescent particles Biophys J 81 4 2378 88 Oct 2001 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 11566807 21450444 0006 3495 Journal Article EMBL CMCI Image Basic Course REFERENCES DW Cromey Digital imaging Ethics 2007 URL http swehsc pharmacy arizona edu exppath D Dormann T Libotte C J Weijer and T Bretschneider Simultaneous quantification of cell motility and protein membrane association using active contours Cell Motil Cytoskeleton 52 4 221 30 Aug 2002 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 12112136 22105498 0886 1544 Journal Article H Geerts M De Brabander R Nuydens S Geuens M Moeremans J De Mey and P Hollenbeck Nanovid tracking a new automatic method for the study of mobility in living cells based on colloidal gold and video microscopy Biophys J 52 5 775 782 Nov 1987 doi 10 1016 50006 3495 87 83271 X URL http dx doi org 10 1016 9 00056 23495 87 83271 X J Gelles B J Schnapp and M P Sheetz Tracking kinesin driven move ments with nanometre scale precision Nature 331 6155 450 3 Feb 4 1988 URL http www ncbi nlm nih gov entrez query fecgi cmd Retrieve amp db PubMed amp dopt Citati
103. image sequence by selecting the Import gt Image Sequence from the File menu ImageJ New Ctrl N Open Ctrl O Open Next Ctri Shift O Open Samples gt Open Recent gt mot o Close crew Ra Save Ctri S pus a Text Image Revert Ciri R TextFile URL Page Setup Print Ctrl P Using QuickTime Video Quit QuickTime Player In the first shown dialog select one of the images in the folder the one you unzipped and click Open Use the second dialog to specify which images in the folder to open and or to have the images converted to 8 bits When working with RGB images like the files of the experimental movie sample it is highly recommended to convert them to 8 bits This way the memory consumption is reduced significantly Check the Convert to 8 bit Grayscale box and click OK Now that the movie is open you can start the plugin by selecting ParticleTracker from the Plugins gt Particle Detector amp Tracker menu File Edit Image Process Analyze Plugins Window Help Eolea a Je amp j2 8 35 y 136 z 0 value 4 Particle Tracker Particle Detection Radius 3 Cutoff 3 0 Percentile 0 10000 96 Preview Detected Save Detected Particle Linking Link Range 2 Displacement 10 00 OK Cancel Second Step select particle detection parameters and preview The dialog showing now has 2 parts Particle Detection and Particle Link
104. img1 img1 img2 Subtract img imgl img2 Multiply img1 img1 img2 Divide img1 img1 img2 AND imgl img1 AND img2 OR img1 img1 OR img2 XOR img1 img1 XOR img2 Min img1 min img1 img2 Max img1 max img1 img2 17 EMBL CMCI Image Basic Course 1 1 Basics of Basics Average img img1 img2 2 Difference img1 img1 img2 Copy imgl img2 aoa As a i Source Destination Add Subtract Multiply Divide Tt FFE MIN Average Difference Copy Note white 0 blackz255 Figure 1 10 Taken from ImageJ web site http rsb info nih gov ij docs menus process html The figure above represents the result of various image math operations Exercise 1 1 5 1 Image Math subtraction Open Images cells ActinDNA tif and cells Actin tif The first image is containing images from two channels One is actin labeled and the other is DNA labeled We isolate the DNA signal out of the first image by image subtraction Do Process Image Calculator In the pop up window choose the appropriate combination to sub tract cells Actin tif from cells ActinDNA tif Don t forget to tick Create New Window 116 RGB image Color images are in RGB format could also be a so called pseudo color image or 8 bit color but this is just because of LUT See section 1 1 7 An other popular format is CMKY but this format is optimized for printing purpose you may have heard it already when printi
105. import the buffer to R You could then plot the profile in those applications Compare original 16bit profile 8 bit profiles with and without scaling by plotting three curves in a graph and discuss the difference Assignments 1 1 4 Simple math on Images 1 Try subtracting certain values from the image you created in the As signment 1 1 1 and check that the values cannot be less than 0 2 Prepare an 8 bit image with pixel value 200 Divide the image by 3 and check the result 3 Prepare a 16 bit image In the File gt New gt Image dialog select 16 bit from the type drop down menu Try adding certain value to check the maximum pixel value 4 Discuss why measurement of fluorescence intensity using digital im age is invalid when some pixels are saturated 50 EMBL CMCI Image Basic Course 1 1 Basics of Basics Assignments 1 1 5 LUT Open Cell Colony tif Use LUT edit function and design your own LUT to highlight the black dots in Green and the background in Black LUT ed itor can be activated by Images gt Color gt Edit LUT Instruction for the LUT editor is at http rsb info nih gov ij plugins lut editor html You might be able to manage using it without reading the web instruction just try LUT lut file could also be edited using Excel Assignments 1 1 6 File size and image bit depth image size If there is an image with width 100 pixels and height 200 pixels what would be th
106. ing The parameters relevant for detection are e Radius Approximate radius of the particles in the images in units of pixels The value should be slightly larger than the visible particle radius but smaller than the smallest inter particle separation e Cutoff The score cut off for the non particle discrimination e Percentile The percentile r that determines which bright pixels are accepted as Particles All local maxima in the upper rth percentile of the image intensity distribution are considered candidate Particles Unit percent 96 In the particle detection part you can play with different parameter and check how well the particles are detected by clicking the preview detected button There are no right and wrong parameters it all depends on the movie type of data and what is looked for Enter these parameters radius 3 cutoff 0 percentile 0 1 default click on preview detected Notice the 2 very close particles marked in the image Check the detected particles at the next frames by using the slider in the dialog menu you cannot move the image itself while the dialog is open With radius of 3 they are rightly detected as 2 separate particles If you have any doubt they are 2 separate particles you can look at the 3rd frame Change the radius to 6 and click the preview button Look at frame 1 With this parameter the algorithm wrongfully detects them as one particle since they are both within the radius of
107. istogram Exercise 1 2 1 3 Local Histogram Equalization Optional Histogram equalization could also be performed on a local basis This becomes powerful as more local details could be contrast enhanced You could try this with the same image g1 tif by Pligins gt CMCICourseModules 54 EMBL CMCI Image Basic Course 1 2 Intensity gt CLAHE if you have installed the course plugin in ImageJ or if you are using Fiji Process gt Enhance Local Contrast CLAHE 1 2 2 Region of Interest ROI To apply certain operation to a specific part of the image you can select a re gion by region of interest ROI tools The shape of the ROI could be var ious such as rectangular elliptical polygon free hand or a line straight segmented or free hand There are several functions that will be used often in association with ROI tools Exercise 1 2 2 1 Cropping Open any image Select a region by rectangular ROI Then image Crop This will remove the unnecessary part of the im age to reduce calculation time Exercise 1 2 2 2 Masking Open any image Select a region by rectangular ROI Then Edit clear Edit Clear Outside After checking what happened do Edit Fill same operation could be done by Edit Selection Crate Mask Exercise 1 2 2 3 Invert ROI Open any image Select a region by rectangular ROI Then Edit Selection Make Inverse In this
108. it depth and the file size I will explain these values later Take the pen tool and draw some shape what ever you want If you do not see anything drawn then you need to change the color of the pen to white by Edit Option Colors and set Fore ground Color to white Then do File gt Save as gt Text image and save the file You will find that the name of the file ends with txt Open File Explorer Win or Finder Mac and double click the file The file will be opened in text editor What you see now in the text editor is a text image a 2D matrix of tab delimited numbers At the left most column in the example Fig 1 3 there are only zeros This corresponds to the left column pixels in the image where the color is black In the middle in the EMBL CMCI Image Basic Course 1 1 Basics of Basics example image there are several 255 These are the white part of the image In the text image edit one of the numbers either 0 or 255 and change to 100 Then save the file with a different name such as temp txt Then in Image open the file by File gt Import gt Text Image You should see some difference in the image now The image now has a dark gray dot not black nor white ipi xd iB xj fle Edt Format View Heb 10x15 pixels 8 bit OK b Figure 1 2 A digital image b is a matrix of numbers b EET igi xi 10x15 pixels 8 bit OK le Edit Format View Help an an
109. jectories are again displayed because by definition every trajectory length is at least 1 spans over at least 2 frames Try other numbers for the filter option and notice the differences Set filter for 100 only 13 trajectories remained after filtering Select the yellow trajectory the one shown here by clicking it once with the mouse left button e All Trajectories Visual Sle 1 144 tutorial0000 228x275 pixels 8 bit 8 amp Results View Preferences Relink Particles Configuration Kernel radius 3 Cutoff radius 0 0 Percentile 0 6 4 Particle Tracker DONE Found 110 Trajectories D trajectories remained after filter 90 trajectories remained after filter 13 trajectories remained after filter FE Filter Options All Trajectories Visualize All Trajectories Focus on Selected Trajectory Save Full Report Selected Trajectory Info Display Full Report A rectangle surrounding the selected trajectory appears on the screen and on the trajectory column of the results window the number 32 is now displayed it indicates the number of this trajectory from the 110 found Now that a specific trajectory is selected you focus on it or get its information Click on Selected Trajectory Info button the information about this trajectory will be displayed in the results window Foc Trajecto Trajectory 32 6 153 989532 132 864670 4 795121 3 200076 8 058751 7
110. k Splitter rteneave Denterte ave Montage to Stack or Stack Maker Ex Start 2 Increment 3 These commands are under Image gt Stack gt Figure 1 29 Editing Stacks 2 Tools Courtesy of Sebasti n Tosi IRB Barcelona 36 EMBL CMCI Image Basic Course 1 1 Basics of Basics e Height 200 Slices 10 Then draw time stamps in each frame by Image gt Stacks gt Time Stamper with the default properties but for the following X location 50 Y location 90 Font Size 36 Now you should have printed time for each frame 0 to 9 sec To make a montage of this image stack do Image Stacks make Montage Set the following parameters and then click OK Columns 5 e Rows 2 e Border Width 1 e Label Slices checked O sec 1 sec 2 sec 5 sec 6 sec 7 sec 8 sec Figure 1 30 Montage of the time stamped stack If you have time try to change the column and row numbers to create montage with different configuration Concatenation Then duplicate the time stamp stack concatenate the duplicate to the original Image gt Stacks gt Tools gt Concatenate Create a montage of this double sized stack 37 EMBL CMCI Image Basic Course 1 1 Basics of Basics 4D stacks A time lapse sequences we could call it 2DT stack or xyt stack or a Z stack we could call it 3D stack or xyz stack are relatively simple objects because they are both made up of a series of two dime
111. kground subtraction for bright field images is available in ImageJ wiki For flat field correction protocol see Optical lnttp www optinav com Polynomial Fit htm lenttp imagejdocu tudor lu imagej documentation wiki how to how to correct background illumination in brightfield microscopy 79 EMBL CMCI Image Basic Course 1 3 Filtering Microscopy Primer site Protocol for the background subtraction for fluorescence images using cal ibration slide and Image could be found in Miura and Rietdorf 2006 For the background removal of fluorescence time series bleaching of flu orescence should also be considered See recent article by Schwarzfischer et al 2011 Deconvolution removes out of focus emission signal See appendix 8 for an extensive tutorial using Image plugin You could also refer to a classic review Wallace et al 2001 1 3 7 Other Functions Fill holes Skeltonize Outline Frequently after some morphological operation we need to fill the holes in a binary image For example we detect the boundary of a cell and want to obtain an object which is filled and covers the cell In this example we will see its effect Exercise 1 3 7 1 Open book text tif Then fill holes by Process Binary Fill Holes Vnttp micro magnet fsu edu primer digitalimaging imageprocessingintro html 80 EMBL CMCI Image Basic Course 1 3 Filtering 1 3 8 Batch Processing Files Once you established a
112. logical Image Processing Mathematical morphology is a powerful tool that can be used to extract features and components from an image Itis often used to pre process or post process images to facilitate a posterior analysis In this process a small shape structuring element not necessarily square like we did in the precious section is translated across the image during the course of pro cessing Certain mathematical logic operations are performed on the image using the structuring element to generate the processed image In this sec tion we first introduce dilation and erosion two fundamental operations in mathematical morphology We then describe morphological operations obtained by combining erosion and dilation Dilation Dilation is an operation that grows objects in a binary image The thicken ing is controlled by a small structuring element In Fig 1 47 you can see 72 EMBL CMCI Image Basic Course 1 3 Filtering the structuring element on the right and the result after applying dilation on a rectangle Erosion Erosion shrinks or thins objects in a binary image After erosion the only pixels that survive are those where the structuring element fits entirely in the foreground Fig 1 48 In above examples the structuring elements are asymmetrically shaped In ImageJ the structuring element used by the morphological filters of the Process gt Filters menu is a square so that the effects are even along both axes Ex
113. ls Then the resulting image becomes 3 x 3 To understand the effect do the following exercise Exercise 1 1 12 1 Open the example image 4pixelimage_sample tif The image is ultra small so zoom it up to the maximum as much as you can you must click on or Ctrl You now see four pixels in the window Dupli cate the image by Image gt Duplicate Magnify again Select all by Edit gt Selection gt Select A11 Then Image gt Adjust gt Size In the dialog window input the width 4 and height 4 corresponds to 200 enlargement Tick aspect ratio and un tick Interpolation Then click OK Check the pixel values in original im age and the enlarged image Exercise 1 1 12 2 Do the similar resampling but this time enlarge the image by 150 Check the pixel values The resampling in the exercise was without interpolation the check box was OFF Interpolation is similar to the one dimensional interpolation we do with graphs In case of images the gradient is also two dimensional so the situation is a bit more complex There are various methods for in terpolating image The interpolation method used in Image is the bilinear 48 EMBL CMCI Image Basic Course 1 1 Basics of Basics EET aloj xi 3x3 pixels 8 hit OK 2x7 pixels 8 hit OF a b Figure 1 37 Artifacts produced by resizing a Four pixel image and b nine pixel image after resizing interpolation Briefly the bilinear
114. lysis of Time Series 1 5 9 ASSIGNMENTS Assignments 1 5 5 2 Open tubulin dynamics image stack kin_tubulin stk and do manual track ing like you did in the exercise 1 5 6 Track six points and plot the tracks in a same graph Is there anything you can say about their dynamics Images courtesy of Puck Ohi 132 References C M Anderson G N Georgiou I E Morrison G V Stevenson and R J Cherry Tracking of cell surface receptors by fluores cence digital imaging microscopy using a charge coupled device camera low density lipoprotein and influenza virus receptor mo bility at 4 degrees c J Cell Sci 101 Pt 2 415 25 Feb 1992 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 1629253 92332618 0021 9533 Journal Article H Berg Random Walk in Biology Princeton University Press Princeton 1993 S Bolte and F P Cordeli res A guided tour into subcellular colocaliza tion analysis in light microscopy J Microsc 224 Pt 3 213 232 Dec 2006 doi 10 1111 j 1365 2818 2006 01706 x URL http dx doi org 10 1111 4 1365 2818 2006 01706 X A E Carlsson A D Shah D Elking T S Karpova and J A Cooper Quantitative analysis of actin patch movement in yeast Biophys J 82 5 2333 43 May 2002 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 11964224 21961476 0006 3495 Journal Art
115. m Image 1 46i release Mar 15 2012 you could check the pixel values by Image gt Transform gt Image to Results This command will send the image matrix shown in Results window as numbers 15 EMBL CMCI Image Basic Course 1 1 Basics of Basics line Since the image bit depth is 8 the available number is only between 0 and 255 When you add 10 to a pixel with its value 255 the calculation returns 255 because the number 265 does not exist in 8 bit world A similar limitation applies to other mathematical operations too If you multiply a pixel value 100 by 3 the answer in normal mathematics is 300 But in 8 bit world that pixel becomes 255 How about division Close the currently working test image and prepare a new image as you did in 1 1 1 Zoom up the image and add 100 Process gt Math gt Add you should see the image turns to gray from black Check pixel values by placing the pointer above the image Now Divide the image by 3 Process gt Math gt Divide Check the result by placing the mouse over the image The pixel value is now 33 Since there is no decimal placeholder in 8 bit the division returns the rounded value of 100 3 33 One could also divide image by any real number such as 3 1415 The answer will be in integer in all cases in 8 bit and 16 bit In case of floating point 32 bit image the calculation results are different We study this in the next exercise Exercise 1 1 4 2 Sim
116. mage Because of the infinitesimally small num bers that can be stored the 32 bit floating point format allows to store a virtually unlimited dynamic range In other words 32 bit floating point images can store a virtually unlimited dy namic range in a relatively compact way with more detail in the EMBL CMCI Image Basic Course 1 1 Basics of Basics shadows than in the highlights and take up only twice the size of 16 bits per channel images saving memory and processing power A higher accuracy format allows for smoother dynamic and tonal range compression Quote from http www dpreview com learn key bits Guide Line for the Choice of Bit Depth In fluorescence microscopy the choice of whether to use higher bit depth format depends on a balance among the intensity of excitation light the emision signal intensity the sensitivity of photon detector quantum effi ciency and the gain If the signal is bright enough then there would good S N that you could simply use a lower bit depth but if the signal is too bright then you might need to use a higher bit depth just to avoid saturat ing the bits This balancing could also be controlled by changing the gain of the camera In the end the choice of bit depth depends on what type of measurement you want to achieve If you only need to determine the Shape of the target object segmentation you might not need a higher bit depth image as you could try to adjust the ima
117. mage once they are converted I recommend not to compress data except for sharing of data for visualiza tion purpose The PNG format does not lose the original pixel values but due to the file format conversion the header information associated with the original will be lost and this often causes problem in retrieving impor tant information about images such as scales and time stamps 1 1 9 Multidimensional data In this section we study the following topics Image stacks Editing Stacks e 4D stacks e Hyperstacks Multidimensional data in which we define here as a set of image data with dimensions more than x and y Multidimensional data have intrinsic lim itation in displaying them on two dimensional screen Efforts have been made to represent multidimensionality in various ways One way is the image stack stacks When you take time lapse sequence or z sections using a commercial mi croscope system the image files are generally saved in the company spe cific file formats e g lsm or lif files Importing these images into ImageJ could simply be done using LOCI bioformats plugin 29 EMBL CMCI Image Basic Course 1 1 Basics of Basics These image data appear as a series of images contained within a window with a scroll bar at the bottom By scrolling one could go through the third dimension like a movie This is the simplest form of multidimensional representation in 2D display In some cases multi
118. n Open rice tif and then select Plugins Segmentation Trainable Segmentation Your task here is to segment rice grains using trainable segmenta tion Zoom up the image using magnifying tool as usual image Then choose free hand ROI tool start marking a rice grain If you are satisfied with ROI then click Add to Class1 Then again using freehand ROI tool mark background Then add this another class by Add to Class2 Your trainable segmentation window should look something like Fig 1 84 Click Train Classifier in the left panel then calculation starts that takes for a while When calculation finishes you will see that most of rice overlaid red and background in green Fig 1 85 Zoom out the image and check again You might see that some of rice grains are not segmented well so we should train more Use freehand ROI tool to mark that was unfortunately categorized as background and add 109 EMBL CMCI Image Basic Course 1 4 Segmentation Training Train classifier Toggle overlay Create result Options Apply classifier Load data Save data Create new class Settings Figure 1 84 Marking signal and background it to class 1 Add to Class 1 Then Train classifier check the segmentation results You could repeat such marking and training until you get a satisfactory segmentation result TIP You could delete specif
119. n this category In the second approach an image is partitioned into regions that are similar according to set of predefined criteria Thresholding and watershed segmentation fall in this category 1 4 1 Thresholding Many biological images comprise of light objects over a constant dark back ground especially those obtained using fluorescence microscopy in such a way that object and background pixels have gray levels grouped into two dominant modes One obvious way to extract the objects from its back ground is to select a threshold T that separates these modes g x y Mad 1 8 0 otherwise Where g x y is the thresholded image of f x y In a sense thresholding is an extreme case of contrast enhancement we studied already 1 2 4 By setting a threshold value T and converting the image pixels with values above T becomes white and otherwise black color could be in inverse 95 EMBL CMCI Image Basic Course 1 4 Segmentation Image thresholding operation turns the image into black and white image which is called binary image Exercise 1 4 1 1 Although there is a function specialized for the thresholding we first try thresholding images using brightness contrast function we used in the previous section Open the image 2D_Gel jpg Then Image gt Adjust gt Brightness Contrast In B amp C contrast con trol window set the minimum and maximum to a same value You then should see the LUT curve is v
120. nce Images In some cases mathematical operations on image sequences is effective in visualizing dynamics Difference images also called subtraction im ages is one of such techniques A difference image of successive frames in a sequence is given by Dj x y IjaGoy IG y 1 10 21For basics to deal with stack go to 1 1 9 115 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series where D is the difference image and j is a given plane in the stack Differ ence images highlight features of the image that change rapidly over time much in the same way that a spatial gradient enhances contrast around an edge The change is usually brought about by movement of an object in the image or by a kinetic process that has not reached steady state like photobleaching Exercise 1 5 1 1 Open image stack 1703 2 3s 20s stk Then duplicate the stack Edit gt Duplicate Don t forget checking the Duplicate entire stack Go back to the original stack delete the last frame frame 31 Image gt Stacks gt Delete Slice Alternatively you could simply click button in the stack tools at the last frame Activate the duplicated stack and delete the first frame Then do subtraction Process gt Image gt Calculator OriginalStack DupicatedStack This will then subtract frame 1 frame2 frame2 frame 3 and so on Do you see Frapped Region in the difference image stack 1 5 2 Projection of Time
121. ng another ROI So the intensity measurement again with this new ROI OPTIONAL Numerical values in the table can be copy and pasted in spread sheet software like Excel or OpenOffice Calc If you know how to use R then you could save the results table as a CSV file and read it from R Try draw a graph in your favorite plotting software When you measure the fluorescence level it is very important to measure the background intensity and subtract the value from the measured flu orescence This is because the baseline level adds offset to the measured value so you could not quantify the true value 1 5 4 Measurement of Movement Dynamics Movement is an essential component of biological system To quantify the movement dynamics various methods has been developed In case of im age sequences particle tracking is a popular way to quantify movement 1 5 6 1 5 7 In any tracking methods the ultimate goal is to obtain the po 117 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series sition coordinate of a target object in each image frame so that the move ment of the target object can be represented as changes in the position co ordinate Velocity and movement direction can then be calculated from the resulting coordinate series Since tracking deals with segmentation and position linking the process is rather complex Besides tracking there is a easier way to represent and quantify movements occurring in sequences The techniq
122. ng of the intensity range while provisioning a sufficient headroom to avoid saturation and as such always the safest option It can be mathematically derived that quantizing a continuous signal to a finite number of levels induces an Additive White un correlated noise commonly called quantization noise The level of this noise source is conditioned by the number of levels that are allowed over the effective range of the signal De pending on the image intrinsic noise shot noise at a given pho ton regime and the noises coming from the detector and the amplifier electronics this noise can eventually be neglected One should always make sure that this noise source can be neglected to avoid trashing valuable available information personal communication S bastien Tosi 11 3 Converting the bit depth of an image In many occasions you might want to decrease the bit depth of image sim ply to reduce the file size 16 bit file becomes half the file size when it is con verted to 8 bit or you might need to use certain algorithm that is available only with 8 bit images there are many such cases or so on In any case this will be a good experience for you to see the limitation of bit depth Here we focus on the conversion of a 16 bit image to an 8 bit image to study its effect and associated possible errors Exercise 1 1 3 1 Let s first open a 16 bit image from the sample If you have the course plugin installed
123. ng something in Pho tolab RGB stands for three primary colors Red Green and Blue If all of them are bright at the same intensity then the color is white or gray 18 EMBL CMCI Image Basic Course 1 1 Basics of Basics If only red is bright then the color is red and so on A single RGB image thus has three different channels In other words three layers of different images are overlaid in a single RGB image Each channel layer has a bit depth of 8 bit So a single RGB image is 24 bit image For this the file size of color pics becomes three times larger than a gray scale 8 bit image Don t save 16bit image in RGB format since you lose a lot of information for automatic conversion from 16 to 8 bit takes place Exercise 1 1 6 1 Working with RGB image a Open the image RGB_cell tif by either EMBL gt Samples or File gt Open Then split the color image to 3 different images Image gt Color gt Split Channels b Merge back the images with Image Color gt Merge Channels leoo Merge Channels C1 red RGB_cell tif red C2 green _RGB_cell tif green C3 blue RGB_cell tif blue C4 gray None C5 cyan None a F C6 magenta None S C7 yellow None V Create composite LJ Keep source images LJ Ignore source LUTs Cancel ox Figure 1 11 Color Merge Dialog In the dialog window choose an image name for
124. ning the peak position xe yc _ Lx C x y T L C x y T _ Ly C x y T Y Eic T 1 16 x 1 15 where T is the threshold value and negative values are discarded xe y corresponds to the centroid of the magnitude of correlation that is thresh olded by T 125 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series Since the cross correlation function 1 14 tends to give higher values at brighter regions rather than at regions of similar shapes a normalized form of cross correlation can also be used Yo ico Ux t iy 7 TK Gj K Cn x y 1 17 1 18 1 19 where I is the mean intensity of a portion of the image overlapping the kernel and Mi and My are the root mean square values of the kernel and the corresponding portion of the image respectively The precision of the measurement is high Cheezum et al 2001 but object tracking will fail when the shape of object changes radically between frames The cross correlation method has been used extensively in single particle tracking SPT SPT is a technique developed for measuring the mobility of membrane bound proteins and the movement of motor proteins with a nanometer precision Gelles et al 1988 Geerts et al 1987 Schnapp et al 1988 Sheetz et al 1989 and was recently reviewed Ritchie and Kusumi 2003 In these studies a single protein was attached to a very small gold particle or labeled with a fluo
125. nlarging When you want to check the details of images zooming is the best way to focus on a specific region to observe details Zooming is done by the magnification tool icon of magnification glass and this simply enlarges or shrinks the pixels Instead if you really need to increase the number of pixels per distance we call such processing as Resampling and we will examine this a bit in this section By the way if we just want to have larger image by adding some margin 47 EMBL CMCI Image Basic Course 1 1 Basics of Basics we call this resizing and the command for this resides in the menu tree under Image gt Adjust gt Canvas Size The resampling changes the original data If we have an image of size 10 pixels by 10 pixels and resample it to 200 the image becomes 20 x 20 If we resample it by 50 then the image becomes 5 x 5 The resampling is a simple task that could be done by Image gt Adjust gt Size This is a simple operation but one must take care about how pixels will be pro duced while enlarging and reduced while shrinking If the enlarging is simply two times larger we could imagine that each pixel will be copied three times to produce a block of four pixels to complete the task The pixel values of the newly inserted pixels will then be identical to the source pixel But what happens if we want to enlarge the image by 150 To simplify the situation think about an image with 2 x 2 pixe
126. nsional images More highly dimensioned data are common these days In many occasions we take a time series of z stacks maybe also with multiple channels The or der of two dimensional images within such multidimensional stacks then becomes an important issue If we take an example of 3D time series a pos sible order of 2D images could start first with z slices and then time series In this case the frame sequence will be something like this imgZO0 T1 tif imgZ1 T1 tif imgZ2 T1 tif imgZ0 T2 tif imgZ1 T2 tif imgZ2 T2 tif imgZ0 T3 tif The number after Z is the slice number and that after T is the file name We often call this order XYZT Alternatively 2D images could be ordered with time points first and then z slices In this case images will be stacked as imgZO T1 tif imgZ0 T2 tif imgZ1 T1 tif imgZ1 T2 tif imgZ2 T1 tif imgZ2 T2 tif We call this order XYTZ The order you will often find is the first one XYZT but in some cases you 38 EMBL CMCI Image Basic Course 1 1 Basics of Basics might also find the second one XYTZ Stacks with more than 4D If we have multiple channels we could even have an order like XYCZT and so on Since when viewed as a regular stack we can scroll only along a single dimension the order of stack dimensions is fundamental as it effects the order in which the images will appear To avoid such complication with dimension ordering there is an advanced format of s
127. o each format as is the size of the header In biology microscope companies create their own formats to include more information about the image such as the type of microscope used used objectives binning shutter speed time intervals user name and so on Having to handle company specific formats makes our life more difficult because each image can a priori only be opened from the software provided 27 EMBL CMCI Image Basic Course 1 1 Basics of Basics by these companies Fortunately there is an excellent ImageJ plugin which enables importing specific image formats to Image You do not have to know all the details about the architecture of various image formats thanks to the bioformats plugin but it is important for you to know that the difference resides mainly in the header The data part is in most cases same something like what we have seen already using text image for more details on header refer to the appendix 1 6 1 Exercise 1 1 8 1 Accessing the image properties Open the example image wt1 tif Do Image gt Show Info Scale pixels inch is listed in the information window which was read out from the header of the image Then do Image gt Properties also showing the scale Compression Image file size become huge as the size of the image become larger There are convenient ways to reduce image file size by data com pression When you take a snap shot using commercial digital camera or smart
128. o more than 100 What is the optimal number of iterations Suggestion you can obtain a quantitative assessment of your prediction as follows keep in mind that this is a synthetic experiment under Log choose Normal under SER choose Process SER and select new reference as the reference image The plugin will indicate the error in dB of the current estimate with respect to the original image What do you observe How can you explain it Perform 50 iterations of the Richardson Lucy algorithm with TV regularization on the image corrupted by Poisson noise with A 0 0005 Comment the result both qualitatively and quantitatively using the same error measurement as before Compare the Richardson Lucy deconvolution results at 50 iterations with and without the TV regularization hint subtract the two images to highlight the differences Exercise 4 3D deconvolution Close all images and the Deconvolution plugin Load the file Microtubules tif into ImageJ this is a widefield stack of a biological sample that we will deconvolve In this exercise we generate a synthetic PSF another option would be to use a PSF obtained from an experimental measurement typically by imaging sub resolution fluorescent beads Launch the PSFGenerator plugin and generate a PSF using the following parameters which correspond to the acquisition settings e ni 1 518 e NA 1 4 e W040 0 e 517nm amplitude 255 background 0 e SNR 100
129. o some experi ments on FFT You could copy amp paste this in a new macro window and install to generate stripes with various frequencies and orientation var llimewicleln var euge 505 macro vertical stripes wieuete Leal SiS p function vertical winsize newlmage stripes 8 bit black winsize winsize 1 setForegroundColor 255 255 255 setLineWidth linewidth for i 0 i getWidth i linewidth 2 drawLine i 0 i getHeight aoa MEerisSsalein JEluhe s s 2 Wreabeinmec V 2 macro diagonal stripes vertical sizex 2 run Rotate angle 45 grid 0 interpolation Bilinear Ug icum exea oo walcleina s sivec Ineielaic srsa wer seals Si wef 2 y raizo 24M n xc bug Cep p MacrostripeGenerator ijm 167 EMBL CMCI Image Basic Course 1 6 Appendices 1 6 10 App 8 Deconvolution Exercise Exercise starts from next page In this exercise following plugins are used If you do not have them in your ImageJ or Fiji download them and install Deconvolution lab cannot be installed by Plugins gt Install Unpack the downloaded zip folder and place all the contents under plugins directory 1 PSFgenerator http bigwww epfl ch algorithms psfgenerator 2 DeconvolutionLab http bigwww epfl ch algorithms deconvolution 168 Excercises on deconvolution Alessandra Griffa September 1 2010 Exercise 1 3D widefield PSF Start t
130. oments Triangle RenyiEntropy Shanbhag Figure 1 74 For testing all algorithms try all choice in AutoThreshold dialog is conve nient The image above is an output of doing this with the image blob tif Algorithm names are printed in small fonts below each image For choosing an algorithm a proper way might be to look for original pa pers describing the algorithm and think which one would work best for your purpose but in practice you could try one by one and just choose one that fits your demand Instead of manually trying out all available al gorithms you could test them in one action by Image gt Adjust gt Auto Threshold and then choose Try All in the dialog window fig 1 74 In batch processing command for automatic threshold would be like this setAutoThreshold Huang dark The first argument is the name of the algorithm and could be any of the 99 EMBL CMCI Image Basic Course 1 4 Segmentation ones listed above 1 4 Feature Extraction Edge Detection Patterns within image have features such as lines and corners To detect them we employ procedure called Feature Extraction In the general sense this means to extract certain object that you are focusing on In a strict definition in image processing this means to do some defined calcu lations with neighboring pixels to get a feature image This output then can be used simply for segmentation but also for machine learning based
131. on amp list uids 3123999 88122616 0028 0836 Journal Article R N Ghosh and W W Webb Automated detection and tracking of individual and clustered cell surface low density lipoprotein receptor molecules Biophys J 66 5 1301 18 May 1994 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 8061186 94339330 0006 3495 Journal Article K Hirschberg C M Miller J Ellenberg J F Presley E D Siggia R D Phair and J Lippincott Schwartz Kinetic analysis of secretory protein traffic and characterization of golgi to plasma membrane transport in termediates in living cells J Cell Biol 143 6 1485 503 Dec 14 1998 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 9852146 99069477 0021 9525 Journal Article 134 EMBL CMCI Image Basic Course REFERENCES A Kusumi Y Sako and M Yamamoto Confined lateral diffu sion of membrane receptors as studied by single particle tracking nanovid microscopy effects of calcium induced differentiation in cultured epithelial cells Biophys J 65 5 2021 40 Nov 1993 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 8298032 94128889 0006 3495 Journal Article D S Martin M B Forstner and J A Kas Apparent subdiffusion inherent to single particle tracking Biophys J
132. on is for two dimensional mobility In case of 3D lt r gt 6Dt 130 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series MSD Time Figure 1 92 MSD plotting and different types of movement a pure random walk b biased random walk diffusion with drifts c constrained random walk d pure random walk but with a lower diffusion coefficient In the case of molecules this could be due to larger molecule size In the case of cells this could be due to a less migration activity e same as b but with a lower diffusion coefficient flow rate the displacement r will be r V4DT 0T 1 25 When MSD lt 7 gt is plotted against time vt causes an upward curvature of the graph For example in case of chemotaxis the direction of movement is biased towards the chemoattractant source so that the curve becomes upward Another mode of movement is constrained diffusion This happens when the diffusion is limited within a space Consider a molecule diffusing in a bounded space The molecule can diffuse normally until it hits the bound ary In such a case the MSD curve displays a plateau such that shown in curve c in Fig 1 92 When T is small MSD is similar to the pure diffusion but as t becomes larger MSD becomes attenuated since the displacement is hindered at a defined distance The constrained diffusion is often observed with membrane proteins Saxton 1997 131 EMBL CMCI Image Basic Course 1 5 Ana
133. oordinates inside the segmented object and is the most commonly used feature for representing the object position The centroid coordinate p x y can be calculated as xyem 1 12 py 2H BH x ny where 91i is the region surrounded by the contour and n is the total number of pixels within that region In other words all x coordinates of the pixels inside the object are added and averaged The same happens for all y co ordinates The derived averages for x and y coordinates represent the cen troid coordinates of the object The position of the object can be measured for every frame of the sequence manually In Image centroid calculation is available in the Particle Analysis function Gaussian Fitting Method For spherical signals such as fluorescence beads or fluorescently labeled sub resolution particle the signal intensity distribution can be fitted to a standard two dimensional Gaussian curve Anderson et al 1992 Sch tz et al 1997 Tardin et al 2003 x Xn Toe Yn Wr I x y zo Znexp 1 13 where I x y is the intensity distribution of an image Zo is the background intensity and z is the height of the peak W is the width of the curve that peaks at Xn Yn This peak position is the signal position Although this fitting method is restricted to spherical or oval objects it yields the most precise measurements even with a very low signal to noise ratio Cheezum 124 EMBL CMCI Image
134. p in the trajectory meaning a part in the trajectory that was interpolated by the algorithm to handle occlusion exit and entry of the particle EMBL CMCI Image Basic Course 1 6 Appendices 1 6 7 App 5 Particle Analysis Copied from ImageJ website Particle Analysis counts and measures objects in binary or thresholded images It works by scanning the image or selection until it finds the edge of an object It then outlines the object using the wand tool measures it using the Measure command fills it to make it invisible then resumes scanning until it reaches the end of the image or selection Press the esc key to abort this process Use Image Adjust Threshold to threshold an image Use the dialog box to configure the particle analyzer Particles outside the range specified in the Size field are ignored Enter a single value in Size and particles smaller than that value are ignored Particles with circularity val ues outside the range specified in the Circularity field are also ignored The formula for circularity is 4pi area perimeter 2 A value of 1 0 indicates a perfect circle Note that the Circularity field was added in Image 1 35e Select Outlines from the Show pop up menu and Image will open a win dow containing numbered outlines of the measured particles Select Masks to display filled outlines of the measured particles or Ellipses to display the best fit ellipse of each measured particles O9 e 950 Gor j
135. p01 then you could place the text exp01 in the text field of File name contains This then avoid loading other data set with prefix exp02 even if they are within the same folder Regular expression could also be used in the or enter pattern text field 7If you do not know what Regular Expression is try the tutorial at http www vogella com articles JavaRegularExpressions article html For those who knows what regex is use the Java regex 31 EMBL CMCI Image Basic Course 1 1 Basics of Basics There are other options such as scaling and conversion of the image bit depth but these operations could be done afterward The im ported image sequence is within one window or a stack A stack could be saved as a single file File extension is typically tif which is same as the single frame file The file header will contain the information on the number of frames that the image contains This number will be automatically detected when the stack is opened next time so that the stack can be correctly reproduced Don t close the stack exercise continues Exercise 1 1 9 2 In the Image tool bar among all the tool icons there is a button with Stk All the commands related to stack can be found there by click ing that icon eoo Blolxio Fiji sja lalalo ail jl a aaa Stacks Menu Figure 1 25 ImageJ tool bar in default mode Start Animation plays the
136. phone saved images are always compressed But keep in your mind There are two types of compression formats loss less and lossy formats In loss less formats pixel values generated by CCD are preserved even after the compression is made On the other hand in lossy formats pixel values become different from the original measured values and cannot be restored PNG is a popular loss less compression format that does NOT alter the original pixel values With this format compression of images are done by shrinking redundant parts for example instead of having 100 pixels of 0 values as a block you could replace that part by saying here there are 100 pixels of zeros If you need to compress files using PNG format is preferred for scientific image data Other more popular compression formats are like JPEG and GIF JPEG is ofen used in commercial diginal cameras In addition to the redundancy shrinking explained above the compression procedure tries to mathemat ically interpolate some parts of the image to ignore small details These 2 LOCI Bioformat Plugin http www loci wisc edu ome formats html 28 EMBL CMCI Image Basic Course 1 1 Basics of Basics are lossy formats as the process of compression discards some part of data This causes artifacts in the image and it could even be manipulation of data For this reason we better avoid using lossy formats for measure ments as we cannot recover the original uncompressed i
137. ple Math on 32 bit Image prepare a new 32 bit image in the New gt Image dialog select 32 bit from the type drop down menu Then add 100 to the image Check that the image pixel values are all 100 Then divide the image by 3 Check the answer This time the result has a decimal placeholder Bit depth limitation of digital image is very important for you to know in terms of quantitative measurements Any measurement must be done knowing the dynamic range of the detection system prior to the measure ment If you are trying to measure the amount of protein and if some of the pixels are saturated then your measurement is invalid 16 EMBL CMCI Image Basic Course 1 1 Basics of Basics 1 1 5 Image Math In the previous section we operated on a single image Likewise we can perform computations involving two images For example assuming two images f and g with same dimensions and f 5 10 100 1 4 5 10 50 1 5 meaning that the pixel value at the position 5 10 in the first image f is 100 and in the second image g is 50 we can add these values and get a new image h such that h 5 10 f 5 10 g 5 10 100 50 150 1 6 This holds true for any pixel at position x y h x y f x y 9 x y 1 7 Note that this only works when the image width and height are identical Above is an example of addition More numerical operations are available in ImageJ Add
138. position alignment Eoy Zero padding ES 0000000100000000 0100000000 k 1 2 3 2 0 oo e e No m 0000000100000000 0000000100000000 D 12320 0 2 3 2 Position after one shift 0000000100000000 0000000100000000 m 3 12320 0 Position after four shifts N 0000000100000000 0000000100000000 n 1 2 3 2 0 0 2 3 2 1 Final position 4 full correlation result full convolution result g 000023210000 000123200000 0 same correlation result same convolution result h 0 0 23 2100 01232000 p Figure 1 44 1 dimensional convolution and correlation figure taken from DIP 64 EMBL CMCI Image Basic Course 1 3 Filtering calculation k Then you multiply each element pairs 5 pairs in this case and sum up the results since all partners in f are 0 the sum of multiplica tion is 0 We note this as the first element of full convolution result 0 We then slide w to the left by one element do the multiplications and summing up again Note the result as the second element of full convolution result o Like wise we do such calculation step by step until last element of w matches the last element of f n After that we throw away padded ele ments from the output 1D array to have a resulting array with same length as the original f p To summarize convolution is implemented sequentially as a local linear combination of the input image using the filter kernel weights We
139. rated to b low frequency part near the origin and c high frequency part in the periphery Above was an example of low pass filtering for isolating low frequency stripes and high pass filtering for isolating high frequency signal We could utilize more complex filters to isolated more specific frequency sig nals Such filter is called band pass filter This is available in Process FFT Band Pass Filter 92 EMBL CMCI Image Basic Course 1 3 Filtering Bos asx BES io x 64x64 pixels 8 bit 4K 50x50 pixels 8 bit 2K Fi b Figure 1 70 a Lower frequency part could then be invert FFT to visualize b only the low frequency pattern Bor o BEES oxi 50x50 pixels 8 bit 2K 64x64 pixels 8 bit 4K Figure 1 71 a Higher frequency part could then be invert FFT to visualize b only the high frequency pattern 93 EMBL CMCI Image Basic Course 1 3 Filtering 1 3 12 ASSIGNMENTS Assignment 1 3 1 Convolution and Kernels 1 Design your own kernel apply it to an image of your choice and dis cuss what it does 2 Gaussian kernel Open the Gaussian kernels in sample image folder Gss5x5 txt Gss7x7 txt and Gss15x15 txt by File gt Import gt Text Image Then try getting the line profile of 2D Gaussian crossing the peak of the curve The line profile across the 2D Gaussian should be 1 D Gaussian curve Save the resulting graphs as image files 3 Visualize the Gaussian k
140. rators such as smooth sharpen find edges and so on Many of them are called linear filters so called because filtering the sum of two equally sized images is equivalent to filtering these images apart and summing the results With non linear filters these two different operations may end up in two different results Smoothening Smoothening operation which is used for attenuating noise Median ker nel is better for shot noise removal but for learning purpose we stick to the smoothing is done by applying the following kernel to the image e ae e BR mi j qi Let s take an example of an image with a vertical line in the middle 0 0 10 0 0 0 0 10 0 0 0 0 10 0 0 0 0 10 0 0 0 0 10 0 0 Just for now we forget about the padding for explanation and first apply the kernel to the top left corner for calculating convolved value at 1 1 pixel position note top left element position is 0 0 Then the calculation is 13This property is absolutely fundamental and at the heart of transform domain based filtering For instance discrete Fourier transform decomposes any discrete signal image in a finite sum of components images knowing the effect of the filter on each of these finite components is hence sufficient to fully characterize the filter and its effect on any signal image 67 EMBL CMCI Image Basic Course 1 3 Filtering output 1 1 0x1 0x1 0x 14 0x1 0x1 0x14 10x1 10x1 4 10x1 9 30 9
141. rophore and its movement was analyzed by video microscopy Theoretical examinations showed that different modes of protein movement can be discriminated with nano meter resolution by measuring the mean square displacement of the labeled proteins Qian et al 1991 Various types of membrane protein motions such as immo bile directed confined tethered normal diffusion and anomalous diffu sion were resolved revealing the kinetics of the membrane protein mobili ties Kusumi et al 1993 Saxton 1997 An automatic tracking program for multiple proteins has been developed by Ghosh and Webb 1994 and was used for measuring the movement of actin patches in Yeast Carlsson et al 2002 126 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series ImageJ Tracking Plugins Here is a list of object tracking plugins available in ImageJ Nov 2006 Particle Tracker http www mosaic ethz ch Downloads ParticleTracker Uses local maxima for estimating particle position and custom criteria for discriminating non particles e MTrack2 http valelab ucsf edu nico IJplugins MTrack2 html e MTrackJ http imagescience bigr nl meijering software mtrackj manual html Initialization should be done manually e Spot Tracker http bigwww epfl ch sage soft spottracker Exercise 1 5 7 1 Open image stack TransportOfEndosomalVirus tif We use Parti cle Tracker for experiencing the automatic tracking Sbalzarini and
142. round subtraction other arguments might also come after depending on which options you selected in the Subtract Background dialog box iBixi Record IMENTE Name Macro ijm Create run Subtract Background rolling 5 disable Figure 1 54 Recorder window after Subtract Background command Copy and paste this text command and paste it somewhere to keep it Then do Process Batch Macro This will createa new window titled batch Process Paste the text command you prepared above in the text field of this batch Process window Fig 1 55 To set the input folder where files to be processed are stored click Input and select the folder of your choice Then set the output folder where processed files will be stored choose one that is empty 82 EMBL CMCI Image Basic Course 1 3 Filtering _ I x J j DA KotaicMChicoursetsample imagestspindle frames Output D _KotatCMCitcoursetsample_imagestout Output Format TIFF Add Macro Code Select from list run Subtract Background rolling 5 disable Test Open Save Process Cancel Figure 1 55 Batch Process window click Output and select a folder Check the options so that they look like above then clicking Process button will start the batch process ing of all files in the input folder Check the images created in the output folder to see images are actually the processed version of in
143. simply by inputting 3 in the Diameter field Click tiles to acti vate deactivate positions Click Apply button to do the actual pro cessing Then you could try with a larger diameter 2 diameter 9 Apply these two different structuring elements to dilate noisy finger 76 EMBL CMCI Image Basic Course 1 3 Filtering print Discuss the difference in outputs AEE Select Operation Dilate binary gray Select Structuring Element circte Diameter 3 E Draw Structuring Element Load Structuring Element Q 255 Binaryze j Info Apply Figure 1 50 Morphological Image Processing dialog to design structuring element 1 3 5 Morphological Image Processing Gray Scale Images The morphological image processing is not limited to binary images Grayscale images can also be the subject In fact the morphological image processing of binary image we have studied so far is a special case of the grayscale processing We now enhance the technique with a slightly more general rule in order to apply the structuring element to grayscale images Erosion we take the minimum value among the pixel positions where the structuring element is overlapping Process gt Filter gt Minimum the grayscale version of open Dilation we take the maximum value among the pixel positions where the structuring element is overlapping Process gt Filter gt Maximum the grayscale ver
144. sion of close The morphological processing of grayscale image is very useful for sub tracting the background or eliminating the shading in an image see figure below One could remove all features smaller than the structuring element 77 EMBL CMCI Image Basic Course 1 3 Filtering by Minimum operation of gray images In the following example we will experience this Exercise 1 3 5 1 Open rice tif Then Process gt Filters gt Minimum In the dia log window input the radius of the structuring element structuring element is circular Adjust the radius to remove the rice grains from the image and only keep the background Then remove the resulting background image from the original image Note For bright field images the background should actually be re moved by division rather than subtraction such as Specimen Darkfield Bright field Darkfield Corrected Image x 255 1 3 6 Background Subtraction Besides the background subtraction we studied above using Minimum and Maximum filtering there is a special function implemented for background subtraction in Image extension of morphological processing This func tion is implemented as the so called Rolling ball algorithm The algo rithm is nothing but the gray scale morphological processing with rolling ball being a circular structuring element but is more convenient as a single command dedicated to the background subtraction In
145. sulting images exhibit high intensity peaks which means higher value that reflect the direction of stripe pattern Fig 1 58 For example FFT image of stripes in horizontal direction Fig 1 57a shows high intensity peaks that are horizontally aligned Fig 1 58a Stripes in vertical direction Fig 1 57c become vertically aligned peaks in FFT image Fig 1 58c In general values in FFT image exhibit peaks aligned in the direction of the repetitive pattern of the original spatial domain image If the pattern does not have preferential direction isotropic pattern such as concentric rings then the FFT image will also be isotropic 18 To generate such stripe images for studying FFT use macro code in Appendix 1 6 9 84 EMBL CMCI Image Basic Course 1 3 Filtering stripes Zaye fed 50x50 pixels 8 bit 2K a Figure 1 57 Original images Spatial domain images stripes ioj x 50x50 pixels 8 bit 2K Y TTC ojx 50x50 pixels 8 bit 2K stripes 2m Sioi x 50x50 pixels 2K b c FFT of En sini fed FFT of En fed FFT of stri PI 64x64 pixels 8 bit 4K 64x64 pixels 8 hit 4K 64x64 pixels 8 bit 4K FFT of stri xl d Figure 1 58 FFT images Frequency domain images High intensity values are high lighted in red 85 EMBL CMCI Image Basic Course 1 3 Filtering Frequency of patterns in spatial domain image also has a clear relationship with t
146. tack called Hyperstack It could have up to three scroll bars at the bottom for channels C slices Z and time points T so one could easily scroll through the dimension of your choice 1 31 We will study more on the actual use of the image stacks in section 1 5 Exercise 1 1 9 2 In this exercise you will learn how to interact with image stacks n Di mensional images Open sample image stack yeastDivision3DT tif This is a 3D time series stack You can browse through the frames by moving the scroll bar at the bottom of the frame Since this is a 3D time series you will see that each time point is a sub stack of images at different optical sections To view the stack in a more convenient way you could convert the stack to a hyperstack by Image gt Hyperstacks gt Stack to Hyperstack Ae In Hyperstack mode two scroll bars appear at the bottom of the window One is for scrolling through z slices and the other to select a time point frame Each scroll bar could be moved independently of the other dimension so you could for example go through the time 8XYCZT isa typical order but many microscope companies and software do not follow this typical order If this conversion does not work properly the first thing you could check is if the setting of the image dimension is properly set To check this Image gt Properties and check if slice z number and frame number time points are set correctly For the sampl
147. tection algorithm due to the set detection parameters and bad quality of the movie did not detect the particle This can also happen in real biological data Since the link range was set to 1 the linking algorithm did not look ahead to frame 72 to check for a possible continuation to the trajectory Re link the particle with link range 2 go to the Relink Particles menu at the results window and select the set new parameters for linking In the dialog now opened set the link range to 2 and click OK When the re linking is done a message will be displayed in the results window Relinking DONE Found 10 Trajectories _ All Trajectories Visual 300 BIizIE3 amp Results 1 100 test linear 03 001 118x118 pixels 16 bit grayscale 2 7M View Preferences Relink Particles 96 Width 118 pixel Height 118 pixel Global minimum 0 0 Global maximum 62 0 4 Relinking DONE Found 10 Trajectories AL Filter Options UN Visualize All Trajectories Focus on Selected Area Focus Save Full Report Selected Trajecti Display Full Report You can already see that fewer trajectories were found 10 instead on 17 Click on the View all Trajectories button and compare the view to the one created with link range 1 Focus on the blue trajectory The previously 2 separate trajectories are now 1 and in frame 71 were the particle was not detected a red line is drawn to indicate a Ga
148. ticle dialog should we change Try again with different parameter values 1 4 5 Machine Learning Trainable Segmentation Segmentation could also be done by letting computer to learn what you think as signal and background To do so one could use machine learning algorithms We use Trainable Segmentation plugin In brief you manually mark what you decide as signal and background and then let the plugin to Train classifier The plugin studies your markings by combining many possible features and it comes up with a model to categorize each pixel into classes to output a segmented image When you start the plugin you will see a window with image you want to segment and buttons in the surrounding Panel in left side is for com mands and the panel in right side is for assigning your markings either signal or background There are only two classes on start up but you could add more classes e g signal 1 signal 2 background by create new class 108 EMBL CMCI Image Basic Course 1 4 Segmentation in the left panel Fig 1 83 Hi Tramable Segmentation Lo 252x252 pixels 8 bit 62K Training Train classifier Toggle overlay Create result Options Apply classifier Load data Save data Create new class Settings Figure 1 83 Trainable Segmentation Window Exercise 1 4 5 1 This exercise is only available with the Fiji distributio
149. tool Width and height of the image are defined by the number of pixels in x and y directions Each pixel has brightness or intensity or more strictly somewhere between black and white represented as a number Within an image file saved in a computer hard disk the intensity value of each pixels are written The value is converted to the grayness of that pixel on monitor screen We usually do not see these values or numbers in the image displayed on monitor but we could access these numbers in the image file by converting the image file to a text file Exercise 1 1 1 1 Conversion of image to a text file Make a new image by File New Image In dialog win dow make a new image with the following parameters e name test txt type 8bit 1 Zooming in out of the image does not change the content of the image It s also possible in a limited way by moving the mouse pointer over the image and checking the number indicated in ImageJ menu bar 4 EMBL CMCI Image Basic Course 1 1 Basics of Basics Fill with Black 10 pixel width 15 pixel height e Slices 1 New Image Name test txt Type 8 bit Fill with Black 1 Width 10 pixels Height 15 pixels Slices 1 Cancel OK Figure 1 1 New Image Dialog Clicking OK you will see a new window showing a black image Fig 1 2 At the top of the window you can see the file dimension 10 x 15 b
150. traction 78 1 3 7 Other Functions Fill holes Skeltonize Outline 80 1 3 8 Batch Processing Files uae ox te as 81 1 3 9 Fast Fourier Transform FFT of Image 83 1 3 10 Frequency domain Convolution 87 1 3 11 Frequency domain Filtering 90 1 3 12 ASSIGNMENTS occu c8 64s eel dt me Pele ee 94 DEPMCTHAHON te 6 vas eana hoo Ste Sore d aw Aes 95 1 4 1 Thresholding ucoksegukRS Parte kare ee Se RA 95 14 2 Feature Extraction Edge Detection 100 14 3 Morphological Watershed 103 1 4 4 Particle Analysis 4 249 a oko RR 105 1 4 5 Machine Learning Trainable Segmentation 108 1 4 6 ASSIGNMENTS 4 8 4 64 e928 64 O48 Ohba X 112 Analysis of Time Series nauau hed aoe Gee IR A 115 1 5 1 Difference Images uer y aa a 115 EMBL CMCI Image Basic Course CONTENTS 1 5 2 Projection of Time Series 2 sui5 ege Yet Yee ee 116 15 3 Measurement of Intensity dynamics 117 1 5 4 Measurement of Movement Dynamics 117 Lb Eynosgraplis 122x195 eels Aol e TRE 118 1 5 6 Manual Tracking 4 229999 R9 RR 121 1 5 7 Automatic Tracking cp Wo ex M 123 15 8 Summarizing the Tracking data 129 15 9 ASSIGNMENTS ecce Bins Ex bee Roe 132 References a quo gp het Gwe de rendir at Bt eoque fete s 133 1 6 Appendic s coc eile ORI ALI ES Re s Ero 139 1 6 1 App 1 Header Structure and Image Files 139 162 App 1 5 Installing Plug In
151. ttp www ncbi nlm nih gov pubmed 11730015 Thomas Walter David W Shattuck Richard Baldock Mark E Bastin Anne E Carpenter Suzanne Duce Jan Ellenberg Adam Fraser Nicholas Hamilton Steve Pieper Mark A Ragan Jurgen E Schneider Pavel Tomancak and Jean Karim H rich Visualization of image data from cells to organisms Nat Methods 7 3 Suppl S26 S41 Mar 2010 doi 10 1038 nmeth 1431 URL http dx doi org 10 1038 nmeth 1431 Jennifer C Waters Accuracy and precision in quantitative fluorescence microscopy J Cell Biol 185 7 1135 1148 Jun 2009 doi 10 1083 jcb 200903097 URL http dx doi org 10 1083 jcb 200903097 C M Witt S Raychaudhuri B Schaefer A K Chakraborty and E A Robey Directed migration of positively selected thymo cytes visualized in real time PLoS Biol 3 6 e160 Jun 2005 URL http www ncbi nlm nih gov entrez query fcgi cmd Retrieve amp db PubMed amp dopt Citation amp list uids 15869324 1545 7885 Journal Article 138 EMBL CMCI Image Basic Course 1 6 Appendices 1 66 Appendices 1 6 01 App 1 Header Structure and Image Files For TIFF format detailed description could be found at http www digitalpreservation gov formats content tiff_ tags shtml 139 EMBL CMCI Image Basic Course 1 6 Appendices 1 6 0 App 1 5 Installing Plug In ImageJ To install Plug In download the Plug In file class or jar and put the file in the plugin folder within Image folder
152. tup please do one of the following two ways ImageJ by ImageMath Each image can be squared by Process gt Math gt Square Then to squared images do Process gt Image Calculator for the addition of two squared images This command pops up a window like below Image Calculator Exi Image microtubule tif x Operation ada gt Image2 microtubule 1 tif x M Create New Window v 32 bit Result Figure 1 75 Image Calculation From the pull down menu select microtubule tif and microtubule 1 tif Don t forget to check create new windowand 32 bit results Then click OK There will be a new window To this new image titled result of microtubule do Process gt Math gt Square Root for the final calculation You prob ably would see only black frame this is because 32 bit im age is not properly scaled and you need to adjust LUT To doso do Image Adjust Brightness and Contrast and 101 EMBL CMCI Image Basic Course 1 4 Segmentation in the Brightness amp Contrast window click Auto to auto scale the LUT Fiji by Image Expression Parser To do complex calculation between two images Image Ex pression parser could be used Select Process gt Image Expression Parser Input following in the Expression field sqrt A 2 B 2 Then Choose microtuble tif for A and microtubule 1 tif duplicate for B B might not be shown at the start up To show B selection field click button at bottom l
153. ue is called kymographs 1 5 5 see below This method loses positional information but for measuring velocity the method is easy and fast The process of tracking has two major steps In the first step the object must be segmented This then enables us to calculate the coordinate of the object position such as its centroid There are various ways to do segmentation see 1 4 In the second step the successive positions of the object must be linked to obtain a series of coordinates of that object We call this process position linking Position linking becomes difficult when there are too many similar objects In this case it would become impossible to identify the object in the next time point among multiple candidates Position link ing also becomes difficult when the successive positions of the object are too far apart This happens when the time interval between frames is too long In this case multiple candidates may appear in the next time point even though similar objects are only sparsely present in the image frame If the target object is single and unique linking of coordinates to successive time points has none of these problems 15 5 Kymographs Kymographs are a two dimensional time traces where time t is in Y axis and space along a one dimensional contour is in X and the dynamical vari able F x t is visualized as an image Kymographs provide a fast and con venient way to visualize motion and dynamics in microscopy images
154. ugh you don t need to downgrade the stack Then do Plugins gt Course gt Manual Tracking a window pops up Fig 1 90 Morak Deeteastpomt Endmak po tet attacks Ej Local maximum E Figure 1 90 Manual Tracker Interface Then 1 Check centering correction use Local Maximum 2 Start manual tracking by clicking Add track 3 End tracking by End Track 4 Show tracks by Drawing Functions 122 EMBL CMCI Image Basic Course 1 5 Analysis of Time Series You could track different particles by repeating steps between 2 and 3 Results window will list the measured positions for particles Fig 1 91a and step 4 will show a track overlay image stack Fig 1 91b File Edit Font 83 121 640x512 pixels RGB 151MB Trackn Slice n X ie Distance Velocity Pixel Value 1 1 1 356 269 1 1 699 2 1 2 355 268 0 182 0 091 924 3 1 3 355 267 0 129 0 065 881 4 1 4 356 267 0 129 0 065 906 5 1 5 357 267 0 129 0 065 946 6 1 6 357 267 0 0 1051 7 1 7 357 268 0 129 0 065 1264 8 1 8 357 267 0 129 0 065 1144 9 1 9 357 266 0 129 0 065 1149 10 1 10 357 265 0 129 0 065 1134 TT 11 355 264 0 288 D 144 1142 12 1 12 354 264 0 129 0 065 749 13 1 13 354 264 0 0 713 14 1 14 355 262 0 288 D 144 796 m 1 1 4 6 0620 11990 n ne TAR m a b Figure 1 91 Manual Tracking Results a Results table and b Track Overlay view
155. way you can select region excluding the region you initially selected Exercise 1 2 2 4 Redirecting ROI Open any two images In one of the image select a region by rectangular ROI Then activate the other image by clicking that window and do Edit gt Selection gt Restore Selection ROI with same size and position will be reproduced in the window Exercise 1 2 2 5 ROI manager You can store the position and size of the ROI in the memory Select a region by rectangular ROI Then Analysis 55 EMBL CMCI Image Basic Course 1 2 Intensity Tools gt Roi Manager Click Add button to store ROI informa tion Stored ROI can be saved as a file and could be loaded again when you restart the Image 12 3 Intensity Measurement As you move the mouse pointer over individual pixels their intensity value are indicated in the Image menu bar This is the easiest way to read pixel intensities but you can only get the values one by one Here we learn a way to get statistical information of a group of pixels within ROI This has more practical usages for research To measure pixel values of a ROI Image has a function Analyze Measure Before using this function you could specify the parameters you want to measure by Analyze Set measurements There are many parame ters in the Set measurement window Details on these parameters are listed in the Appendix 1 6 4 For intensity measurements following par
156. zation and local histogram normalization If the histogram is occupying only part of the available dynamic range of image bit depth 8 bit 0 255 16 bit 0 65535 we could adjust pixel val ues of image to increase its range so that contrast become more enhanced There are two ways to do this normalization and equalization Normalization With normalization pixel values are normalized according to the minimum and the maximum pixel values in the image and bit depth If the minimum pixel value is pmin and the maximum is pmax in an 8 bit image then nor malization is done as follows OriginalPixel Value pmin NewPixel Value l pmax pmin 255 Equalization Equalization converts pixel values so that the values are distributed evenly within the dynamic range Instead of describing this in detail using math formula I will explain it with a simple example see Fig 1 38 We con sider an image with its pixel value ranging between 0 and 9 We plot the histogram from this image the result looks like in the figure 1 38a For equalization a cumulative plot is first prepared from such histogram This cumulative plot is computed by progressively integrating the histogram values see Fig 1 38b black curve To equalize flatten the pixel intensity distribution within the range 0 9 we would ideally require a straight diag onal cumulative plot in other words the probability density of the pixel intensity distr

Basics of Image Processing and Analysis

Contents

Download Pdf Manuals

Related Search

Related Contents