BraTumIA 1.2 handbook
Contents
1. BraltumlA A software tool for automatic Brain Tumor Image Analysis Handbook Version 1 2 30 October 2013 developed by Stefan Bauer Thomas Fejes Raphael Meier Mauricio Reyes Institute for Surgical Technology and Biomechanics University of Bern Switzerland stefan bauer istb unibe ch in collaboration with Johannes Slotboom Nicole Porz Alessia Pica Roland Wiest Departments of Neuroradiology and Radio Oncology Inselspital Bern University Hospital Switzerland WINSELSPITAL UNIVERSITATSSPITAL BERN HOPITAL UNIVERSITAIRE DE BERNE BERN UNIVERSITY HOSPITAL UNIVERSITAT BERN 1 2 3 o N OA WO A Contents OVO OW aoia E cates e esa vaneteeee a Gadeva ed aside dete asenseceecad 3 System Requirements and INStallatiO Ms casia E asl ae oe 3 Ser INAV ealhede escececst cciaaceate eed alc ee cecil ces ven de ach Oral danas eed cia anereh Oem eaisb canta ue dceee acs ass 4 opel The man WWII OW cccicasesiev alba diesiahic duierda stv uct eas aicopeail Mec EE A N 4 3 2 HE ACMI UNCC el U aea shcece eaten casecaceatectt a aaa 4 3 3 PROCESSING the datda E N iter sorleni be ead seicareusweeeaeeds 5 3 4 Visualizing and analyzing the results ccccccsscccccssecccceescccceeseccceeececseesecesseecesseneceesuusecessuneses 6 3 5 Locating the data on disk for external USE tics tnicediadens davectertiecsndowiesedindena dare teseenibeneaee 8 KNOWMISSUES esac acetate iether cece ah Outs eae auc eat cuss OA 9 SCICMEI2. MulihtodalReg Classification AIGOnce Apol for all Serie Segmentation Table Patier 1 1 Series 1 1 5 J Reading finished 3 4 Visualizing and analyzing the results The visualization sub windows can be used in two ways when all is ticked in the modality selection panel on the left then all four modalities are displayed in clock wise direction starting with T1 on the top left Tlcontrast top right T2 bottom right and FLAIR bottom left If only one specific modality is ticked then this modality is displayed in axial coronal and sagittal view in 3 sub windows and the fourth sub window shows the Ticontrast image Wi Bratu nA oOo r j Load Unload Surgery Date Info Interface i Patient Nam Age Sex G 25 68 Acquisition Da 12 09 2008 Modality i Ti Register d Classifier MultiSere Segmentation See SEG Do SEG Segmentation Table Patier 1 1 Series 1 1 lt 8 Reading finished In the panel on the left it is also possible to choose whether the registered version or the original version of the images should be displayed The Tissues option allows the user to choose if the segmented label image should be shown as a color overlay over the grayscale images and the Structures options will show the subcortical Structures as an overlay In the Show Seg menu the user can choose which modalities should be overlaid with the color label map Finally when th
3. atlas to the patient image Finally the label maps can be transformed back into the original space of each image sequence so that they can be shown as an overlay on the original images The computation time for the complete pipeline depends on the processor and the amount of memory on modern computers it should take around 5 minutes to process one patient Please be aware that BraTumlIA is intended for scientific use only 2 System Requirements and Installation The program runs on Windows 7 or Windows 8 64bit version with a minimum of 16 GB of RAM memory more is recommended Start the Windows installer program and choose a location to install the program on your computer or extract the files which are in the zip folder into a directory of your choice approximately 2 GB of hard disk space are required 3 User Manual The core program is organized as a pipeline where skull stripping multi modal registration tissue classification and segmentation of subcortical structures are carried out sequentially From the main window the user has to load the image data first then process it before he can visualize and analyze the results 3 1 The main window The main window of BraTumlA contains menu buttons for loading and unloading images at the top iii E a Waal Load Unload Surgery Date Info Interface Patient Nam Age Sex G 25 68 Acquisition Da 12 09 2008 Modality Tl ALL Flair Registert Classif
4. automatically extracted from the segmentation result The subcortical structures which have been segmented can be identified by the numbers they are assigned with The subcorticalStructures txt file in the structures folder provides information which structure is represented by which number The originalFiles folder contains the original input files T1 Tlcontrast T2 FLAIR template in mha format For research purposes this format can be handled more easily than Dicom stacks The Registration folder contains all image sequences after registration and skull stripping in mha format It also contains the transformation parameters which have been used for each sequence in the tfm files The SkullStripped folder contains the brain mask of the Ticontrast image plus the skull stripped Ticontrast image in mha format 4 Known Issues e Possible problems with tumors at the skull border due to skull stripping inaccuracies e Possible problems with pediatric patients e Not optimized for post operative images yet e In general the algorithm tends to over segment tumor more false positives than false negatives especially edema in infratentorial regions e Better performance on high grade gliomas than low grade gliomas e The accuracy of the subcortical structure segmentation has not been carefully evaluated yet 5 Scientific Background Brain tumor segmentation is a difficult task and despite a vast amount of scientific literature 1 there are
5. 2003 S Bauer L P Nolte and M Reyes Fully automatic segmentation of brain tumor images using support vector machine classification in combination with hierarchical conditional random field regularization in MICCAI International Conference on Medical Image Computing and Computer Assisted Intervention 2011 vol 14 no Pt 3 pp 354 61 S Bauer T Fejes J Slotboom R Wiest L P Nolte and M Reyes Segmentation of Brain Tumor Images Based on Integrated Hierarchical Classification and Regularization in MICCAI BraTS Workshop 2012 R Meier S Bauer J Slotboom R Wiest and M Reyes A Hybrid Model for Multimodal Brain Tumor Segmentation in MICCAI BraTS Workshop 2013 B Menze A Jakab S Bauer J Kalpathy Cramer K Farahani J Kirby Y Burren N Porz J Slotboom R Wiest L Lanczi E Gerstner M A Weber T Arbel B Avants N Ayache P Buendia L Collins N Cordier J Corso A Criminisi T Das H Delingette C Demiralp C Durst 11 16 17 18 M Dojat S Doyle J Festa F Forbes E Geremia B Glocker P Golland X Guo A Hamamci K lftekharuddin R Jena N John E Konukoglu D Lashkari J Antonio Mariz R Meier S Pereira D Precup S Price T Riklin Raviv S Reza M Ryan L Schwartz H C Shin J Shotton C Silva N Sousa N Subbanna G Szekely T Taylor O Thomas N Tustison G Unal F Vasseur M Wintermark D Hye Ye L Zhao B Z
6. PIG Baek Ch OUI T sic6s 240th ce dole E a N A tate stk dante ceeeasinateaiaeea lesa oboe leeeiake 9 HOWTO CEBIA ssc ecco castes edees Lacndececetactet gh cna E N ewteec neni siete aed 10 PICKMOW CGP CIMEIALS aein a T N a 10 RETE rONCE nana E A E E NER 11 1 Overview BralumlA is a software tool for automatic brain tumor image analysis It can segment the tumor including its sub compartments from magnetic resonance images MRI of glioma patients For this it requires four different MRI sequences T1 Ticontrast T2 FLAIR as an input and it outputs volumetric information about the tumor and its sub compartments necrotic tissue active enhancing tumor tissue non enhancing tumor tissue and edema Additionally the software can also segment healthy subcortical structures surrounding the tumor Label maps of the segmented tissues and structures are available as an overlay on the original images The images are processed using a pipeline approach where skull stripping is performed first in order to generate a brain mask Subsequently all images are co registered to ensure voxel to voxel correspondence between the different MRI sequences Based on these registered images a segmentation of the patient images into healthy and tumor tissues is done based on combined classification and regularization This produces a label map and quantitative information about tissue volumes Healthy subcortical structures are segmented using a deformable registration of an
7. Wen D R Macdonald D a Reardon T F Cloughesy a G Sorensen E Galanis J Degroot W Wick M R Gilbert A B Lassman C Tsien T Mikkelsen E T Wong M C Chamberlain R Stupp K R Lamborn M a Vogelbaum M J van den Bent and S M Chang Updated response assessment criteria for high grade gliomas response assessment in neuro oncology working group J Clin Oncol vol 28 no 11 pp 1963 72 Apr 2010 N Porz S Bauer A Pica P Schucht J Beck R K Verma J Slotboom M Reyes and R Wiest Multi Modal Glioblastoma Segmentation Man versus Machine PLoS One vol 9 no 5 p e96873 May 2014 T Fejes S Bauer J Slotboom R Wiest and M Reyes A Framework for Medical Image Analysis of Brain Tumors in Annual Meeting of the Swiss Society for Biomedical Engineering 2012 L Ibanez W Schroeder L Ng J Cates and others The ITK software guide 2nd ed Kitware 2005 Kitware VTK User s Guide 11th editi Kitware Inc 2010 Qt Online Available http qt project org S Bauer T Fejes and M Reyes A Skull Stripping Filter for ITK 2012 J B A Maintz and M A Viergever A survey of medical image registration Med Image Anal vol 2 no 1 pp 1 36 Mar 1998 J P W Pluim J B A Maintz and M a Viergever Mutual information based registration of medical images a survey IEEE Trans Med Imaging vol 22 no 8 pp 986 1004 Aug
8. e multi scroll box is ticked it is possible to scroll through all modalities simultaneously this makes only sense for the registered images otherwise only the sub window under the cursor is active 6 The segmentation table is a pop up window that shows the volumes for each segmented tissue compartment From this window it is also possible to switch on and off certain layers of the color label overlay by simply clicking on the respective tissue T Color Code Tissue Ty Data Post OP Tin Not set Acquisition Da Not set Tissue Types Volumes in cn White Matter 411 156 7 605 4 212 26 7 95 313 When the mouse cursor is located above a sub window additional activities are possible Scrolling through the slices can be achieved with the scrolling wheel of the mouse or the scroll bar on the right When pressing the left mouse button and moving to the left and right or up and down windowing and contrast of the image can be adjusted Zooming is possible by pressing the right mouse button and moving up or down The keyboard shortcut shift r can be used to reset the current sub window By pressing the Create Report File button a csv file is created that contains a longitudinal volumetric analysis of the active patient It is stored in the output folder of the first baseline scan Interface Patient Nam Age Sex GBM_O4 12 Femak Acquisition Da 19 02 2011 Modality Ti Tic ALL Flair T Regi i Tissues Multi
9. folder LOAD Surgery Da e g 21 12 2013 and press ENTER In addition to the four MRI sequences it is possible to specify a template image to which all other images will be registered This can be e g a CT of the same patient so that the segmentation can be overlaid on the patient CT image or an MRI atlas image for normalization in a standard space In longitudinal patient studies it can also be useful to use the Ticontrast image of the first baseline scan of this patient as a reference template for all acquisition time points in order to ensure voxel to voxel correspondence and allow for direct comparisons across different time points If a template for spatial normalization is not required then this field can be left empty Finally an output folder has to be selected where all the results will be stored 3 3 Processing the data For the processing the user can call each module separately skull interface stripping multi modal registration classification or he can press the Patient Nam Age Sex l ae G 25 G8 All Once button for a completely automatic processing in the Do Seg tab Aciedinition Da 1209 2004 located on the left panel The classification button and also the All Once Modality button will perform both segmentation of tissues and subcortical T1 Tic structures The progress can be seen in the command line window Ai if Registers Classiie vf Malise Segmentathan See SEG Do SEG okullStrippang
10. hao D Zikic M Prastawa M Reyes and K Van Leemput The Multimodal Brain Tumor Image Segmentation Benchmark BRATS S Bauer H Lu C P May L P Nolte P Buchler and M Reyes Integrated segmentation of brain tumor images for radiotherapy and neurosurgery nt J Imaging Syst Technol vol 23 no 1 pp 59 63 Mar 2013 T Rohlfing N M Zahr E V Sullivan and A Pfefferbaum The SRI24 multichannel atlas of normal adult human brain structure Hum Brain Mapp vol 31 no 5 pp 798 819 May 2010 D a Gutman L a D Cooper S N Hwang C a Holder J Gao T D Aurora W D Dunn L Scarpace T Mikkelsen R Jain M Wintermark M Jilwan P Raghavan E Huang R J Clifford P Mongkolwat V Kleper J Freymann J Kirby P O Zinn C Moreno C Jaffe R Colen D L Rubin J Saltz A Flanders and D J Brat MR Imaging Predictors of Molecular Profile and Survival Multi institutional Study of the TCGA Glioblastoma Data Set Radiology pp 1 10 Feb 2013 12
11. ier v MultiSerc Segmentation SeeSEG Do SEG SkullStripping __MultiModalReq Classification Al Once Apoly for all Seria Segmentation Table Patier 1 1 Series 1 1 E GD amp Reading finished On the left panel the user can find different options for processing and visualizing the data This panel also contains basic patient information information about the currently ongoing actions and quantitative information about the segmented tumor volumes The largest part of the main window contains four visualization sub windows where the image data and the label overlay are shown The different sub windows can either show different MRI sequences or one sequence in different viewing orientations 3 2 Loading the data The user can start to load the patient images by clicking on the Load button and add the sequences He has to choose the location of the T1 T1contrast T2 and FLAIR images BraTumlA supports volumetric images in the meta image format mha the Nifti format nii and Dicom series dcm If the format is Dicom the different modalities have to be located in different folders and the location of the folder for each modality has to be selected The directory tree has to be selected in the left sub window and the final folder image in the right sub window Set Modalities Set T1 path Set Tic path Tic Set T2 path T2 Set Flair path Flair Same as Tic Template Output Folder Set Output
12. only very few tools that can be used in a clinical context BraTumlA is a research tool that aims at bridging this gap between science and clinics For this it builds on experience that was previously gained with the Doctor No suite 2 It allows clinicians to perform tumor segmentation and volumetry as suggested by the RANO group 3 in addition to segmentation of subcortical structures BraTumIA has recently been evaluated in a clinical study 4 The complete functionality is integrated into a graphical user interface 5 that is easy to handle for radiologists All the computation and visualization algorithms are implemented in C using ITK 6 VTK 7 and Qt 8 The skull stripping is based on 9 whereas the registration is a standard rigid registration with mutual information similarity metric which can handle different modalities 10 11 The core of the program is the tissue segmentation method which is based on a machine learning approach for integrated classification and regularization 12 13 14 The segmentation method which is integrated in BraTumIA won a Kitware award at the MICCAI BraTS 2012 challenge and an award from the National Cancer Institute at the MICCAI BraTS 2013 challenge for being among the best performing methods for brain tumor segmentation 15 The segmentation of subcortical structures is based on a simplified version of 16 where an atlas 17 is non rigidly registered to the patient image in contrast
13. se O Structuri Segmentation See SEG Do SEG Tl Tile Tz wf Flaar Segmentation Table i Create Report File Patier 1 1 Series 1 1 he j Reading nnished Now reading Series L 3 14 21107 3 99 2 155 9 000001 207141040254 3700004173 Reading finished How reading Series 1 3 12 2 1107 5 99 2 325 9 2000001 202141040254 TA Eeti If the show template window button is clicked the template image is shown in a separate pop up window The template image can also be overlaid with the segmentation results of tissues or structures if the relevant buttons are ticked 3 5 Locating the data on disk for external use All the results are stored in a folder structure which is located in the output directory that has been chosen by the user when loading the data The classification folder and the structures folder contain the label image in the registered image space plus the label images which have been transformed back into the original space of each modality T1 Ticontrast T2 FLAIR template in mha format If the input images were in Dicom format this folder also contains all the label output images in Dicom format so that they can be uploaded to a PACS system for further use Additionally the classification folder contains a report file in txt format listing the segmentation volumes for all the healthy and pathologic tissue sub compartments and also the convential 2D RANO diameter measurements which are
14. to 16 no tumor growth model is used here The definition of the four different tumor sub compartments follows the the VASARI guidelines of the National Cancer Institute of the American NIH 18 7 http www2 imm dtu dk projects BRATS2012 http martinos org qtim miccai2013 6 Howto cite BraTumlA If you use BraTumlA please cite the following publication N Porz S Bauer A Pica P Schucht J Beck R K Verma J Slotboom M Reyes and R Wiest Multi Modal Glioblastoma Segmentation Man versus Machine PLoS One vol 9 no 5 p e96873 May 2014 This handbook can be referred to as S Bauer T Fejes R Meier M Reyes J Slotboom N Porz A Pica and R Wiest BraTumlA A software tool for automatic Brain Tumor Image Analysis 2013 7 Acknowledgements The development of BraTumIA was funded by grants from different institutions e EU projects ContraCancrum and CHIC e Swiss Institute for Computer Assisted Surgery SICAS e Bernese Cancer League e Swiss Cancer League e Swiss National Science Foundation 10 8 References 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 S Bauer R Wiest L P Nolte and M Reyes A survey of MRI based medical image analysis for brain tumor studies Phys Med Biol vol 58 no 13 pp R97 R129 Jul 2013 S Bauer and M Reyes Doctor No A Suite of DoctorEye Plugins by UniBe 2011 P Y
Download Pdf Manuals
Related Search