Home

MARS: Multiple Atlases Robust Segmentation

1. Currently we support 4 label fusion methods in MARS as shown in the droplist box in Fig 6 Majority voting denoted as MV is the baseline label fusion method since it only uses the label information and label the target image by following the majority labels The advantage is that the computation speed is very fast Local Weighted Voting denoted as LWV determines the label at each voxel by compute the patch similarity between each atlas image patch and the target image patch where the patch center is actually the voxel under consideration Thus the vote for each atlas patch has the weight which 1s related with the patch similarity In principle if two image patches look similar they should bear the same labels Non local Mean goes one step further that allow to search in a certain neighborhood in order to alleviate the possible registration error The last label fusion method is called UNC IDEA SuperMAS which is developed in UNC IDEA lab Compared to the above three label fusion methods we use the sparsity constraint multi scale feature representation label specific patch partition and hierarchical label propagation techniques Methods v Majority Voting g Local Weighted Voting Non oca Mean UNC IDEA SuperMAS qp wewer T Fig 6 Currently supported label fusion methods in MARS To set the parameters please click Setting button in the main GUI In the parameter setting dialog t
2. 15 41 0 6698 JU sers grwu eee ta na01_label har 16 42 0 6587 17 43 0 7032 Methods 18 44 0 7909 Majority Voting z 19 45 0 6747 Setting Run 20 46 0 6508 e P Check Rsult Exit Add Clear Load Save z o Label propagation is done Fig 10 The dice ratio in each ROI 4 Contact us For any questions or bug reports please email to grwu med unc edu 5 References 1 Aljabar P et al Multi atlas based segmentation of brain images Atlas selection and its effect on accuracy Neurolmage 2009 46 3 p 726 738 2 Hsu Y Y et al Comparison of Automated and Manual MRI Volumetry of Hippocampus in Normal Aging and Dementia Journal of Magnetic Resonance Imaging 2002 16 p 305 310 Liu J and J Ye Moreau Yosida regularization for grouped tree structure learning in Advances in Neural Information Processing Systems 2010 Li Y et al Discriminant analysis of longitudinal cortical thickness changes in Alzheimer s disease using dynamic and network features Neurobiology of Aging 2012 33 415 427 Wu G et al Robust Patch Based Multi Atlas Labeling by Joint Sparsity Regularization in MICCAI Workshop on Sparsity Techniques in Medical Imaging 2012 Nice France Wu G et al A Generative Probability Model of Joint Label Fusion for Multi Atlas Based Brain Segmentation Medical Image Analysis 2013 In press
3. C IDEA SuperMAS 3 4 Check and save the labeling result After setting the parameters the users can click Run button to perform the multi atlas segmentation It is worth noting our software does not provide the function of image registration Therefore please use other image registration software e g HAMMER SUITE https www nitrc org projects hammer_suite to register all atlas images to the target image in order to have good label fusion performance After clicking Run button the status of label fusion will be shown in the bottom of Main GUI and a message noticing that the label fusion is done will be pop up as shown in Fig 8 Atlas Information A aimed da tekade i Intensity Image 1 Users grwu Projects MARS TestData na02_cbq hdr Users grwu F Input Subject Users grwu Projects MARS TestDa 3 Users grwu Projects MARS TestData na04_cbq hdr Users grwu F ta na01_cbq hdr 2 Users grwu Projects MARS TestData na03_cbq hdr Users grwu F 4 Users grwu Projects MARS TestData na05_cbq hdr Users grwu F V Ground Truth 5 Users grw e wu F Users grwu Projects MARS TestDa E ta na01_label hdr 6 Users gr Label Propagation is done twu F 7 Users gn A Wwu F Methods 8 Usersigm Wwu F Majority Voting la Setting Run m 1 z Check Rsult Exit Add Clear Load Save Label propagation is done Fig 8 The GUI after successfully running the label fusion To check the result user ca
4. ESS FOR A PARTICULAR PURPOSE 2 System and Installation This software is developed using QT http qt nokia com and has been tested on Windows XP 32 bit 64 bit Window 7 32 bit 64 bit MAC 64 bit OS X 10 9 2 and Linux 64 bit kernel version 2 6 18 194 el5 Since GLIRT is for the large data analysis we suggest the computation machine has more than 2G memory and 10G free disk space 2 1 Overview of MARS In general MARS has integrated several state of the art multi atlas based segmentation methods such as majority voting local weighted voting and non local patch based segmentation methods Specifically we also included our recently developed patch based segmentation method in this software Compared with convention methods our method has the following advantages 1 add sparsity constraint to suppress the influence of misleading patches 2 reduce the joint risk of two patches jointly making the same segmentation errors 3 augment the conventional image patch with multi scale representation and label specific patch partition and 4 use iterative framework to correct the possible mis segmentations In order to efficiently segment the ROIs in imaging based researches we developed user friend GUI to facilitate the fast and easy processing of hundreds of images 2 2 Installation for Windows User For windows user you only need to extract the zip file MARS zip to a specific folder and double click the binary file MARS e
5. Software Release 1 0 1 Last updated April 30 2014 MARS Multiple Atlases Robust Segmentation Guorong Wu Minjeong Kim Gerard Sanroma and Dinggang Shen erwu mjkim gerard sanroma dgshen med unc edu Image Display Enhancement and Analysis IDEA Laboratory Department of Radiology and Biomedical Research Imaging Center BRIC University of North Carolina at Chapel Hill Chapel Hill NC 27599 1 Purpose A lot of medical imaging based studies demand accurate segmentation of anatomical structures in order to quantitatively measure structure differences across individuals or between two groups 1 2 For example multiple brain regions need to be automatically delineated for hundreds of brain MR images before constructing brain connectivity network 3 4 To this end automatic ROI Region of Interest labeling has been a hot topic in medical image processing areas as evidenced by many labeling and label fusion methods that have been developed to improve both segmentation accuracy and robustness The software is hosted at IDEA lab webpage of University of North Carolina at Chapel Hill http bric unc edu ideagroup free softwares and NITRC http www nitrc org projects mars Our software is free to use for the academic research purpose Please cite the paper 5 6 as references if your studies use this software This software is distributed WITHOUT ANY WARRANTY without even the implied warranty of MERCHANTABILITY or FITN
6. he text file which encodes above atlas information After click ok button in the load atlas dialog we will see the atlas information as shown in Fig 5 Mars Atlas Information Total atlas number 8 Intensity Image 1 Users grwu Projects MARS TestData na02_cbq hdr Users grwu F Input Subject 2 Users grwu Projects MARS TestData na03_cbq hdr Users grwu F a 3 Users grwu Projects MARS TestData na04_cbq hdr Users grwu F 4 Users grwu Projects MARS TestData na05_cbq hdr Users grwu F Ground Truth on Users grwu Projects MARS TestData na06_cbq hdr Users grwu F D gt Users grwu Projects MARS TestData na07_cbq hdr Users grwu F Users grwu Projects MARS TestData na08_cbq hdr Users grwu F Methods 8 Users grwu Projects MARS TestData na09_cbq hdr Users grwu F Majority Voting Setting Run Check Result Exit Add Clear Load Save Load the atlases from file Fig 5 Atlas information by loading from the text file 3 2 Load the target image 66 The user can load the to be labeled target image by click the button in the right side of Input Subject in the main GUI If the user have the ground truth of target image and want the software to run the analysis program for quantitative evaluation please check the Ground Truth button and load the label of target image by click the button beside Ground Truth 3 3 Set the label fusion parameters
7. he user will the following parameters as show in Fig 7 The explanation and the suggested value for each parameter is conclude in Table 1 Specifically if the user is only able to do the linear registration we suggest using large search radius e g 4 voxel points If the user can do the deformable image registration and the registration results are good enough the search radius can be reduced to 1 voxel point If the user want to fast label fusion one option is to use high pre selection level which can discard more image patches than low pre selection level 800 Dialog Image Patch Radius 3 Pre Selection Level 1 0 69 in each dimension low high Search Radius 2 Sigma Value as 7 0650 low high v Patch Normilzation Sparseness Degree _ J 002 low high Co Fig 7 The GUI for parameter setting Table 1 The specification of the parameters used in MARS Name Meaning Value Suggested Range Value Patch Size The radius of image patch 1 10 2 Search Radius The radius of searching neighborhood 1 10 2 Patch Normalization Indicate whether we need to normalize the intensity true false true values in each image patch Pre Selection Level The threshold to discard the less similar image patches 0 1 0 8 Sigma Value The sigma value in exponential penalty 0 001 0 1 0 01 Sparseness Degree The strength of sparsity constraint 0 001 0 05 0 01 2 applicable to LWV 3 applicable to Nonlocal Mean 4 applicable to UN
8. n click the Check Result button in the main GUI Then please proceed following steps to check the label fusion result 1 Save the label fusion result e g Result hdr as displayed in Fig 9 Currently we only support the 8 bit ANALYSIS format 2 The software will automatically compute the Dice ratio in each ROI and the overall Dice ratio across all ROIs and reported in the main GUI as shown in Fig 10 The user can use the scroll bar to browse the quantitative results 3 The user also can save the Dice ratio result in a text file by clicking the Save button in the main GUI 0 0 8 Save Labeling Result Save As Result nar J Tags alaale iy im amp 6 Desktop Q FAVORITES Fl png J Dropbox I MICCAI i i R21HD081467 01 PDF J Applications Screen Shot t 2 56 46 PM Jesktop Screen Shot t 3 05 24 PM k Documents Screen Shot t 3 08 38 PM Screen Shot t 3 14 00 PM Downloads Screen Shot t 3 23 13 PM H Movies Screen Shot t 3 29 53 PM J Music Screen Shot t 3 30 01 PM cones Screen Shot t 3 42 14 PM Screen Shot t 3 49 04 PM Qi grwu New Folder Cancel Save Fig 9 Save the label fusion result of the target image e090 Mars Atlas Information Total atlas number 8 Label Dice Ratio 11 31 0 6495 Input Subject 12 32 0 6627 Users grwu Projects MARS TestDa 13 33 0 656 ta na01_cbq hdr 14 34 0 5232 V Ground Truth se
9. round Truth Methods Majority Voting psd Setting Run Check Rsult Exit Add Clear Load Save Adding atlas and label map Fig 4 Atlas information after you load a pair of intensity image and label image Another way is to load a set of atlas images from a text file In the testing data we provided a sample of text files that specify the file names of all intensity images and label images Suppose you have 8 atlases which include 8 intensity images and 8 label images As the text file shown below each line tells the software the file names of intensity image and label image 1 Users grwu Projects MARS TestData na02_cbq hdr Users grwu Projects MARS TestData na02_ label hdr 2 Users grwu Projects MARS TestData na03_cbq hdr Users grwu Projects MARS TestData na03_label hdr 3 Users grwu Projects MARS TestData na04_cbq hdr Users grwu Projects MARS TestData na04_ label hdr 4 Users grwu Projects MARS TestData na05_cbq hdr Users grwu Projects MARS TestData na05_label hdr 5 Users grwu Projects MARS TestData na06_cbq hdr Users grwu Projects MARS TestData na06_ label hdr 6 Users grwu Projects MARS TestData na07_cbq hdr Users grwu Projects MARS TestData na07_label hdr 7 Users grwu Projects MARS TestData na08_cbq hdr Users grwu Projects MARS TestData na08_label hdr 8 Users grwu Projects MARS TestData na09_cbq hdr Users grwu Projects MARS TestData na09_ label hdr Click the load button in the main GUI and choose t
10. ve MARS Mulitiple Atlas Robust Segmentation IDEA Lab UNC CHAPEL HILL Guorong Wu 2014 1 28 Fig 2 Welcome screen of MARS in Linux Mac 3 The MARS GUI The main GUI is shown in Fig Windows and Fig Linux Next we will illustrate how to use our software to label ROIs in your data step by step 3 1 Load the atlases and label maps Suppose you have a set of atlas intensity images and their label maps In MARS you can load the atlases and label maps in two ways First you can input a pair of intensity and label images at a time by click Add button in main GUI Then you will see the open file dialog as shown in Fig 3 In current version we only support 8 bit ANALYZE format Please select the header file with file extension hdr for either intensity image and label image After that click ok in the open file dialog Then you will see the selected files in the main GUI as shown in Fig 4 You may drag the horizontal scroll bar to check the atlas information You should repeat this steps until you have loaded all pairs of intensity image and label image To delete particular atlas please click particular atlas and click the clear button in the main GUI 0 Oo Dialog Label Image ok Cance Fig 3 Load the intensity image and label image for each atlas e090 Mars Atlas Information a ee l Intensity Image 1 Users grwu Projects MARS TestData na02_cbq hdr Users grwu F Input Subject G
11. xe Then you will see the GUI as show in Fig 1 8090 Mars Atlas Information Total atlas number 0 Intensity Image Label Image Input Subject Ground Truth Methods Majority Voting Ls Setting Run Check Result Exit Add gt l Clear Load l Save MARS Mulitiple Atlas Robust Segmentation IDEA Lab UNC CHAPEL HILL Guorong Wu 2014 1 28 Fig 1 Welcome screen of GLIRT in window XP 2 3 Installation for Linux Mac User For linux mac user please download the software MARS tar gz and extract using command tar xzfv MARS tar gz to certain folder for example home usrname MARS If you are using C Shell you can set up the path by typing setenv PATH PATH home usrname MARS your c shell profile usrname cshrc If you are using BASH type export PATH PATH home usrname MARS in your bash profile usrname bashrc Then use the source command to make the change take effect immediately source usrname cshrc for C Shell user and source usrname bashrce for BASH user To test if you have successfully set the MARS software you can type MARS in terminal window If you can see the welcome screen shown in Fig congratulations you are able label your medical image data by MARS e800 Mars Atlas Information Total atlas number 0 Intensity Image Label Image Input Subject Ground Truth Methods Majority Voting usd Setting Run Check Rsult Exit Add Clear Load Sa

MARS: Multiple Atlases Robust Segmentation

Contents

Download Pdf Manuals

Related Search

Related Contents