SepINRIA: A Free Software to analyze Multiple Sclerosis Brain MRI

Transcription

1 INRIA Sophia-Antipolis Research Project Asclepios SepINRIA: A Free Software to analyze Multiple Sclerosis Brain MRI Tutorial for SepINRIA v1.7.0 Project Management: Jean-Christophe Souplet (Jean-Christophe.Souplet@sophia.inria.fr) Scientific Direction: Christine Lebrun, Grégoire Malandain, Jean-Christophe Souplet Documentation writers: Pierre Fillard, Jean-Christophe Souplet, Nicolas Toussaint INRIA Sophia Antipolis - Research Project ASCLEPIOS. 2004, route des Lucioles - BP Sophia Antipolis Cedex France

2 Contents 1 Introduction 3 2 Image importation, management and visualization Database General settings New exam importation Image Management Visualization Functions presentation 12 4 Lesion Segmentation Editor, LSE Introduction Applications Segmentation modes Display Image importation, management and visualization LSE Tabs Processing Tab Image Tabs Delineation Tab How to segment in LSE? Manual Segmentation Semi-automatic segmentation Segmentation principle What are LSE results? Lesions in 3D Label File Statistics Conclusions on LSE Automatic Lesion Segmentation, ALS Introduction Applications Method boundaries Display Image importation, management and visualization ALS Principles Presentation

3 5.3.2 Requirements Data processing What are ALS results? Conclusions on ALS Image or Segmentation Comparison, ISC Introduction Applications Segmentation comparison criteria Display Image importation, management and visualization Images Loading Segmentation definition Load images To align (To register) What are LSC results? Image difference To get statistics Conclusions on LSC SepINRIA algorithms and references 37 8 About 39 2

4 Chapter 1 Introduction SepINRIA is a free software dedicated to Multiple Sclerosis (MS) patient brain MRI analysis. It aims at providing to clinicians tools allowing to analyze MS brain images: ˆ easy to use interface thanks to a database, ˆ visualization of images in 2D and 3D, ˆ comparison of different sequences previously aligned together, ˆ quantitative lesion burden evaluation (manually or automatically), ˆ comparison between a segmentation and a segmentation of reference. SepINRIA website (Figure 1.1) can be found at: Figure 1.1: SepINRIA website SepINRIA is available for Windows XP/Vista, Linux Fedora Core and MacOSX. SepIN- RIA is based on ITK, VTK, wxwidgets, vtkinria3d libraries and the MedINRIA framework (Figure 1.2). Figure 1.2: SepINRIA platforms, libraries and framework 3

5 Chapter 2 Image importation, management and visualization Contents 2.1 Database General settings New exam importation Image Management Visualization Database SepINRIA uses a database to store images and to write results. Database management panel is available on the left of the SepINRIA application and is showed in Figure General settings General Settings allow to change the main directory of the database or the directory containing the atlases (this last directory should be set correctly at the SepINRIA installation process). Note: The database is organized this way: directory A directory by patient ID A directory by exam date Data files in home made format New exam importation Images can be imported from a DICOM folder or from other formats. The image file formats (corresponding to the second choice) include those of ITK (Insight ToolKit): Analyze 7.5 (.hdr,.img), Metafile (.mha,.mhd), Gipl (.gipl), VTK image data (.vtk), Dicom (.dcm), Ge4x, GDCM, Nrrd, Siemens Vision, PNG, BMP, TIFF, JPEG,... For dicom images without extensions, please use All filter. After you selected the format of your data, press the import button. The importation of your files will be now realized in 3 steps: 1. selection of the file(s), 2. verification that each image are in the radiological convention, 4

6 Figure 2.1: SepINRIA database panel 5

7 3. verification and/or specification of Patient ID, Exam Date and image type for each image. Selection of the file(s) If you choose the DICOM folder option: Figure 2.2: The DICOM importer main window, shown with brain exam loaded. Click Open button to set the directory to scan. The DICOM importer is a simple wizard that allow the user to import images directly from a DICOM exam. It can reconstruct 3D volumes from 2D DICOM files. IMPORTANT : This tool does not import 3D DICOM files! You can open them with the option from other format files. First click on the Open button to choose the DICOM root directory that contains the exam. Careful: the importer will scan recursively into this directory, don t choose a directory that contain several exams to prevent from memory overloads. The volumes should now appear in the main table window. Each line with an arrow represents a volume. Double click on it to see what files this volume is made from. Clicking on one file will popup the corresponding image on the view window in the upper right corner. The files should be shown in correct order, following a consistent strategy : files are first splited in series and ordered by the image position DICOM flag. If for some reason the user ends up with a mixed up volume, it might be because of image position conflicts (It often happen in T2/proton density protocols). Then the user can click on split button that provides a consistent split process among the image positions given in the DICOM files. Clicking on reset button will go back to the original configuration. 6

8 Last step (click finish ) will reconstruct 3D volumes from the last given fileset configuration. If you choose the from other format files option: a selection box is displayed where you can browse into folders and select files. Multiple files selection is allowed. Note: If you choose multiple files, you will iterate the two next steps (Orientation, Specification) for each file. Verification of image orientation SepINRIA algorithms need to know images orientation. It has been fixed that the images have to be saved into the database in radiological convention. However DICOM files or other format images can be in other conventions. So please verify the orientation of each file that you import into the database. Figure 2.3: The Reorientation Tool main window. This tool helps the user to manually re-orientate the image if this one doesn t match the radiological conventions correctly. The 3 views correspond respectively to the axial view, the coronal view and the sagittal view of the image. If the image has been acquired in another direction than axial, you can switch the acquisition orientation flag in the check box. If the image is still flipped over an axis, you can use switch the flipping flags for the three directions. I-S corresponds to Inferior-Superior direction, R-L to Right-Left, and A-P to Anterior-Posterior. Once the image meets correctly the conventions yon can click on OK. Specifications On this window you can chose the patient id (has to be without space), the date of exam, and the type of the image. If you import images from the DICOM format the different flag can be already filled with corresponding values. You are still in charge to verify them. You have to click finish when all the informations are filled correctly. 7

9 Figure 2.4: The Image specifications tool window Image Management You can select an image (Patient ID, date of exam (YYYYMMDD), image type). Then you can change its orientation thank to the rotate button or export it to standard files formats. 2.2 Visualization In SepINRIA. images are usually displayed in a main window with 4 views (cf Figure 2.5). Figure 2.5: Visualisation of an image in SepINRIA Three 2D views are shown in the main window, corresponding to the axial, coronal and sagittal views of the volume. On each view some information is shown, such as the resolution of the slice (in pixels), the size of a voxel (in mm), the index of the slice being shown, and the value of the pixel at the current selection). By default, a trilinear interpolation is made between pixels. Press i to disable this interpolation. 8

10 Figure 2.6: 2D view navigation. On the left you see a 2D view (axial) centered on the screen. When the zoom interaction is selected, you can zoom in the view by moving the mouse up while left-clicking on the view and translate the view by middle-clicking (middle figure). You can disable the interpolation between pixel by pressing i (right figure). Reset the image position by pressing r. Several mouse interactions are available. With the selection interaction (black arrow), you will navigate in the volume (i.e. select a slice) by moving the mouse up or down while left-clicking on a view. Press r to reset the position to default. The second interaction (windowing) controls the brightness/contrast of the image by moving the mouse while left-clicking on the view. A left-right movement controls contrast while up-down controls brightness. Note that with these two interactions, the 4 different views are synchronized. Press r to reset the contrast to default value. With the zoom interaction, you can zoom in or out the 2D view by left-clicking and moving the mouse up or down. A middle click in the view will translate the image. Press r to reset the zooming. Keyboard and mouse on 2D screen : - Press i to activate or deactivate interpolation between pixels. - When selection interaction is ON (black arrow), move the mouse up or down while left clicking to change slice. This action can also be done using keys and. - When windowing interaction is ON, move the mouse left/right to change the contrast, move the mouse up/down to change brightness. - When zoom interaction is ON, move the mouse up/down while left clicking to zoom. Move the mouse while middle clicking to translate the view. - For any interaction, press r to reset the interaction. The different 2D views can be shown in fullscreen with a click on one of the buttons which are represented in figure 2.7. Figure 2.7: Buttons to put one of the 2D views into full screen. 9

11 On the lower-right part of the main window is shown a 3-dimentional representation of the volume. You can manipulate the volume in different ways: ˆ Rotate the volume by moving the mouse while left clicking in the 3D view. ˆ Translate the volume by moving the mouse while middle-clicking in the 3D view. ˆ Zoom into the volume by moving the mouse up or down while right-clicking in the 3D view. You also can display the 3D view in full screen by pressing the full screen button. When the 3D view is in full screen, you can use the snap shot button to easily save a.jpg picture of the current screen. Figure 2.8: 3D view. You can see on the left a Multi-Planar Reconstruction (MPR) of the image. On the right you see a Volume Rendering (VR) of the image, cropped by the cropping box to visualize inside the volume. The cropping box can be manipulated by control points (red arrow). You can choose between displaying an image with Multi-Planar Reconstruction (MPR) (Fig. 2.8, right) or with Volume Rendering (VR) (Fig. 2.8, right). When VR mode is chosen, you have the possibility to take out a part of the volume in order to visualize inside it. This can be done with the cropping box. Use the control points around the box (red arrow in Fig. 2.8, right) to resize it and crop the volume (left-click on them). The control point in the center of the box allows you to translate it. When you have finished cropping the volume, you can make the box disappear by typing b on your keyboard. The orientation cube and the 3D axes help to recognize the current orientation of the image, and can be switched on or off with i keyboard key. Keyboard and mouse on 3D screen: In the 3D view, several optional features are available : - Shift+left-click translates the volume. 10

12 - Ctrl+left-click rotates the volume around the axis perpendicular to the screen. - Press j activates joystick mode (continuous movement mode). - Press t disables the joystick mode. - Shift+left-click on the cropping box translates it. - Right-click on the cropping box makes it grow or shrink. - Press r to center the image. - If you cannot access the cropping box : Press b to switch it on, then press the center view button. If you still don t see it, it may be inside the volume. Then translate it outside the volume. - If the cropping box doesn t seem to work properly, it might be because it has been flipped over. Use successively the control points of the box to flip it back to normal. 11

13 Chapter 3 Functions presentation SepINRIA is divided in different functions. A function can be loaded/unloaded from the menu Functions thefunction. When a function is loaded, it will be displayed in the SepINRIA main window, replacing any other previously loaded function. This is the list of SepINRIA functions: ˆ Lesion Segmentation Edition (LSE), manual or semi-automatic segmentation of MS lesions (e.g. a segmentation realized by an expert). Segmentations can be saved and visualized in 2D or 3D. Lesion number and lesion volume can be computed and print. ˆ Automatic Lesion Segmentation (ALS), automatic segmentation of MS lesions from four MRI sequences (Dual Spin Echo T2-PD, T1, T2-FLAIR). ˆ Image or Segmentation Comparison (ISC), registration (alignement) of two images, computation of the image of difference to assess evolution and comparison between a segmentation (e.g. automatic segmentation) and a segmentation of reference (segmentation of an expert). In all functions, images (DICOM or not) can be imported into the database and visualized in 2D and/or 3D. Other modules will be available soon: ˆ Brain Atrophy Evaluation (e.g. linear measurements) 12

14 Chapter 4 Lesion Segmentation Editor, LSE Contents 4.1 Introduction Applications Segmentation modes Display Image importation, management and visualization LSE Tabs Processing Tab Image Tabs Delineation Tab How to segment in LSE? Manual Segmentation Semi-automatic segmentation Segmentation principle What are LSE results? Lesions in 3D Label File Statistics Conclusions on LSE Introduction Lesion Segmentation Edition is a function of the SepINRIA software. SepINRIA is dedicated to Multiple Sclerosis patient brain MRI analysis and can be downloaded from : The aims of LSE are: ˆ to simplify the segmentation of lesions on brain conventional MRI, ˆ to give segmentation statistics (Number of Lesions, Volume of Lesions). 13

15 4.1.1 Applications Lesion segmentation is a mandatory step in Multiple Sclerosis (MS) patients diagnosis. However a precise manual segmentation is a fastidious task and is subject to inter and intra expert variabilities. LSE has been developed, tested and validated on brain conventional MRI of MS patients but it may also be used to segment lesions on MRI of elderly or epileptic or diabetic subjects or of patients with cardiovascular risk factors Segmentation modes Up to now, two segmentation modes are available: ˆ manual: The user segments lesion by lesion and slice by slice, ˆ semi-automatic: The user segments in 3D, lesion by lesion thanks to a clic in the middle of the lesion Display Figure 4.1 shows a snapshot of the LSE loaded into the application main window. Figure 4.1: The LSE window loaded into the main application window. 4.2 Image importation, management and visualization Please refer to the chapter 2. 14

16 4.3 LSE Tabs In the LSE, the data visualization is divided in Tabs. Figure 4.2 shows the different tabs in the case where three images (Test PD.inr.gz, Test T1.inr.gz, Test T2.inr.gz) have been loaded. The first tab, called Delineation is used to segment the images. Other tabs allow to select the loaded images that you want to watch. An other tab on the left, called Processing (cf. Figure 4.3), allows to select the image and/or the lesion mask which will be loaded. Figure 4.2: Tabs in LSE Processing Tab Figure 4.3: Processing tab in LSE. The selected images (Patient ID, Date of exam, image type) can be loaded thanks to the Open button. Note: When an image has been loaded the following images to be loaded have to share the same size and voxel size. If they do not, you can align (register) the different available MRI sequences of a same exam time point thanks to the registration button (cf. Figure 4.4). When an image is open and its image tab is displayed, you can load the mask which is selected on this image thanks to the ROI load button. This mask has to share the same size and voxel size with the image. 15

17 Figure 4.4: Button to align the different MRI sequences of a same exam time point Image Tabs When a file is open, it is shown in a image tab. Figure 4.5 shown the content of Test T2.inr.gz tab. Figure 4.5: Test T2.inr.gz tab. Different interactions are possible. Please refer to section Delineation Tab If you clic on the Delineation tab (first tab in figure 4.5), you will see a screen similar to figure 4.6. Images display The tab is divided in two parts. On the left, a 2D view will allow to segment images. On the right, there is a 2D view for each loaded images. The different view are synchronized (all the different view show the same brain slice). As for the image tab you can select a 2D slice, change the window level or zoom thanks to the same buttons. The following buttons (figure 4.7) which were used to put in fullscreen a 2D view in the image tab, are now used to change the slice orientation. 4.4 How to segment in LSE? As we saw, the Delineation has to be realize in the Delineation tab (see 4.3.3). First, an image has to be loaded in the 2D view on the left thanks to the button at the bottom of the 16

18 Figure 4.6: Delineation tab with 3 images. Figure 4.7: Buttons to change the slice orientation in Delineation tab. smaller images on the right (see figure 4.8). The 2D view on the left will be used to segment the lesions. The views on the right are available to help the segmentation. Figure 4.8: Buttons to export the view to the 2D delineation view on the left. 17

19 4.4.1 Manual Segmentation Lesions can be segmented manually. This part of the documentation is a copy of the corresponding part of DTItrack manual. This segmentation procedure has to be realized on the 2D delineation view on the left. Select a slice where you would like to start drawing a lesion (to change slice, you can press keys and ). You can zoom in for more precision. Now choose the ROI color in the color selector. Note that the label 0 (black) color will erase a region (make it as no-roi). Press j to activate the Lesion segmentation mode, you will see a purple cross somewhere in the view. Maintain a left-click and move the mouse to define the contour of the ROI for this slice. The contour is shown in green. You must maintain your click until you reach the first point of the contour (the purple cross). You can also use an alternative mode: by middle-clicking, you can define several control points to draw the ROI contour. When you have enough control points, Ctrl+middle-click on the first control point to finish the drawing. You can then right-click on one of these control points to move it. You can add a control point by pressing Shift+right-click between two existing control points. You can delete a control point by pressing Ctrl+right-click on it. Once you have finished with the contour drawing, you must validate it by pressing v, you will see that the region you just define is now filled by the color you chose in the color selector. - Lesion segmentation mode tips - As said above, this mode can be activated and deactivated by pressing j. When it is activated, no interaction is possible in the 2D view, but you can still change slice by pressing or. - See figure 4.9 for a step by step illustration Semi-automatic segmentation The other way to segment lesion is semi-automatic. First, you have to click in the middle of the lesion (Be careful to be in select a 2D slice interaction mode). Then, you select the label color. Then you have to press the Region Growing button (see figure 4.10). A 3D region will appear quickly with the same color than the one you choose. See figure 4.11 for a step by step procedure description. This segmentation method is still underdevelopment and may not work perfectly at the beginning (see next section to know how to improve the obtained segmentation) Segmentation principle In the LSE, the Delineation Tab is used to segment lesions on any of the loaded images. When a lesion is segmented (manually or semi-automatically), the obtained segmentation has 18

20 Put a 2D view in full screen and zoom in to the region of interest. Choose a color in the color selector. Press j to activate the ROI generation mode. A purple cross will appear on the view. Mode 1: Left-click and maintain to freely draw the ROI contour. Release the mouse only when you reach the first cross. Mode 2: Middle-click to draw the ROI with control points. Ctrl+middle-click on the first cross to finish the contour. Right click on a point to move it. Shift+right-click between points to add one. Ctrl+right-click on a point to delete it. Once your contour drawn, Press v to validate it. The region is then colorized. Use Label 0 color (black) to erase a region. Figure 4.9: Step by step procedure for ROI generation. Repeat this procedure for several slices until you are satisfied with the ROI. This illustration has been realized for the DTITrack (a part of MedINRIA software) manual but the procedure to segment a lesion is similar. 19

21 Figure 4.10: Region growing button. to be validated on the correct image. First, select on which image the segmentation has to be applied (button on the smaller images bottom). Then press the validate roi button. See figure 4.12 for a step by step procedure description. If you go back to the Delineation tab, you can reset the ROI on all the images of the Delineation tab thanks to the reset ROIs button. ROIs which have been previously validated are not reseted in the image tab. You can also realize a new segmentation that you can validated on another image or add to the PD image segmentation. Warning: If you validate the new segmentation on the PD image, you erase the former one. In the Delineation tab you can also reload a segmentation from an image tab thanks to the ROI reload button. The ROI validation, the ROI addition and the ROI reloading are realized on the image which is selected in the delineation tab. Application: The region growing segmentation did not work perfectly on a lesion (see figure 4.13). Moreover, the obtained segmentation can also be reduce thanks to the manual segmentation (label 0 = eraser). 4.5 What are LSE results? Lesions in 3D When a segmentation has been validated on a image, you can see the segmentation in 3D by pressing the Lesion 3D button in the corresponding image tab. Figure 4.14 show an example of this 3D representation Label File When a segmentation has been validated on an image, you can save it from the image tab into the database. A new mask of the lesion will be available in the list of the lesion mask. You can export this mask from the database thanks to the export function of the SepINRIA database panel. In the future, you will be able to reload this file in the LSE to modified it or to load it in other functions of SepINRIA Statistics In the image tab you can press on the roi statistics button. A window similar to figure 20

22 Clic in the middle of a lesion that you want to segment (Be careful to be in select a 2D slice interaction mode). Select the label color and press the Region Growing button. A 3D regions will appear on all the images (you can change of slice to see the 3D region). Figure 4.11: Step by step procedure for Region Growing segmentation. 21

23 The lesion has been segmented on the T2 image but I want to associate it with the PD image. 1) I select the PD image. 2) I press the validate roi button. The segmentation is in the PD image tab. Figure 4.12: Step by step procedure for ROI validation. 22

24 1) I click in the middle of one lesion. 2) After a click on the region growing button, I obtain a first unperfect segmentation of the lesion. I validate it on the T2 image. 3) I click on a part of the lesion which was not correctly segmented. Then I click on the region growing button. I add this new ROI to the T2 image. 4) I reload the ROI from the T2 image. The segmentation is acceptable. Figure 4.13: Region growing correction. Figure 4.14: 3D representation of a segmented lesion. 23

25 4.15 will be displayed. This statistic report is printable. Figure 4.15: ROI statistics. 24

26 4.6 Conclusions on LSE LSE increases segmentation precision, is less time consuming than current manual delineation, allows to visualized in 2D and 3D the segmented lesions, to save the segmentations and gives statistics on segmented lesions. 25

27 Chapter 5 Automatic Lesion Segmentation, ALS Contents 5.1 Introduction Applications Method boundaries Display Image importation, management and visualization ALS Principles Presentation Requirements Data processing What are ALS results? Conclusions on ALS Introduction Automatic Lesion Segmentation (ALSC) is a function of the SepINRIA software. SepINRIA is dedicated to Multiple Sclerosis patient brain MRI analysis and can be downloaded from : The aim of ALS is to perform automatically a segmentation of MS lesions from four MRI sequences (FAST SPIN ECHO T2-PD, T1, T2-FLAIR) Applications Lesion segmentation is a mandatory step in Multiple Sclerosis (MS) patients diagnosis. However a precise manual segmentation is a fastidious task and is subject to inter and intra expert variabilities. That s why a fully automatic method to segment MS lesion like ALS is needed. 26

28 5.1.2 Method boundaries ˆ The method needs the four MRI sequences. ˆ The method does not segment oedematous transient lesions and lesions of the cerebellum or of the brain stem Display Figure 5.1 shows a snapshot of ALS loaded into the main application window. Figure 5.1: The ALS window loaded into the main application window. 27

29 5.2 Image importation, management and visualization Please refer to the chapter ALS Principles Presentation ALS is divided in two parts: ˆ In the process tab (cf. Figure 5.2) you can choose to display images if they are available in the database. ˆ In the toolbar (cf. Figure 5.3), you can process images to get the segmentation of the lesions. Figure 5.2: The ALS process tab. Figure 5.3: The ALS Toolbar Requirements To process images it is mandatory that the 4 sequences (T2, PD, T1 and T2-FLAIR) are available in the database. Moreover the T2 and PD have to be already aligned (registered) (Dual Fast Spin Echo acquisition). 28

30 5.4 Data processing The first time that you use ALS, it is recommended to perform this workflow step by step and to verify intermediate files thanks to the process tab. ALS needs 4 sequences (T2, PD, T1, T2-FLAIR) to perform the lesion segmentation. The T2 and PD sequences are supposed to be already aligned (Dual Fast Spin Echo acquisition). In ALS workflow, other sequences have to be aligned on the T2 thanks to this button. When the registration is done you will be able to display T1 and T2-FLAIR sequences aligned on the T2 from the processing tab. Press this button to perform the brain mask. In MRI images are bias (two voxels with the same biological composition may not have the same intensity). Press this button to unbias the sequences. ALS workflow needs a segmentation of the different brain compartments. Press this button to perform theses segmentations. Press this button to perform the lesion segmentation. Note: If the workflow works correctly on your data, it s not necessary to follow this step by step process. You can press directly the Lesion segmentation button. 5.5 What are ALS results? ALS performs a binary segmentation of the MS lesions. This segmentation, called Automatic lesion segmentation, can be used in other SepINRIA module or exported. For example, the segmentation can be loaded on the T2 sequence in the SepINRIA function: Lesion Segmentation Edition (LSE) (cf. Figure 5.4) and statistic can be obtained (cf. Figure 5.5). Figure 5.4: Left: T2 sequence displayed in LSE; Right: The same sequence with the segmentation obtained by ALS 29

31 Figure 5.5: Statistics (obtained with LSE) of An Automatic Lesion Segmentation (obtained with ALS). 5.6 Conclusions on ALS The aim of this module is to perform automatically a segmentation of Multiple Sclerosis Lesion from four sequences (T2, PD, T1, T2-FLAIR). Future work will allow segmentation of oedematous transient lesions. 30

32 Chapter 6 Image or Segmentation Comparison, ISC Contents 6.1 Introduction Applications Segmentation comparison criteria Display Image importation, management and visualization Images Loading Segmentation definition Load images To align (To register) What are LSC results? Image difference To get statistics Conclusions on LSC Introduction Image or Segmentation Comparison (ISC) is a function of SepINRIA software. SepINRIA is dedicated to Multiple Sclerosis patient brain MRI analysis and can be downloaded from : The aims of ISC are: ˆ to assess difference between two images. MRI sequences can be automatically aligned (co-register). The image of difference can be computed. ˆ to compare quantitatively two segmentations. Automatic computation of comparing criteria. 31

33 6.1.1 Applications Neurologists need to compare MRI sequences (e.g. to assess evolution). Because of patient motion between the different scans or because of image resolutions, a same voxel in the two images can correspond to a different brain location. The images have to be aligned (registered) before comparing them. ISC allow to realize this step and to compute an image of difference. Lesion segmentation is a mandatory step in Multiple Sclerosis (MS) patients diagnosis. A lot of different semi-automatic or automatic methods are available in the literature. However these methods are not always validated with the same criteria. Consequently, to compare these methods can be a fastidious task. ISC aim to compute statistics between a segmentation of reference and a tested segmentation. The segmentation of reference (SegRef) can be obtained thanks to an expert using the function: Lesion Segmentation Edition of SepINRIA. The tested segmentation (SegTes) can be the one obtained by an automatic method as the function Automatic Lesion Segmentation of SepINRIA Segmentation comparison criteria The validation criteria are computed in two steps. ˆ The lesion detection step. ˆ The lesion delineation step. For the moment and for each of this steps, the following criteria are computed: ˆ Sensitivity, ˆ Positive predictive value Display Figure 6.1 shows a snapshot of LSC loaded into the main application window. 32

34 Figure 6.1: The LSC window loaded into the main application window. 6.2 Image importation, management and visualization Please refer to the chapter Images Loading Segmentation definition A segmentation is a binary file where the pixel is equal to 1 if it correspond to a lesion or equal to 0 in the other case Load images Images or Segmentations selections are realized thanks to the Processing panel on the left of the application (cf Figure 6.2). After the selection of both images/segmentations, you can load them. 6.4 To align (To register) When the two images/segmentations have been loaded, you can press the registration button. If the images do not share the same size and voxel size, the registration is performed 33

35 Figure 6.2: Processing panel in LSC. otherwise a window asking if the image are already aligned is displayed. If you choose YES the Comparison Tab (cf Figure 6.3) is displayed otherwise the registration is perform first. Figure 6.3: Comparison tab in LSC. 34

36 6.5 What are LSC results? Image difference The image difference can be computed (e.g. We can see the lesion which is not in both segmentation; Figure 6.4). Figure 6.4: Image difference. Note: The image of difference is the absolute value of the difference between the two images To get statistics As soon as you have loaded the two segmentations, the button Get statistics is enabled and you can press it. After a moment, results will be displayed in a frame similar to the one in figure

37 6.6 Conclusions on LSC Figure 6.5: ROI statistics. The aims of this function are to compare two images or segmentations. First the images can be aligned to asses evolution between different MRI sequences or sequences from different exam time point. Then the image of difference can be computed. Statistics to compare two segmentations can also be performed. 36

38 Chapter 7 SepINRIA algorithms and references If you use SepINRIA for any purpose (e.g. publication), a reference to the program and the authors must be included. A list of publication related to SepINRIA can be found at This is a part of these publications: SepINRIA: Jean-Christophe Souplet, Christine Lebrun, Pierre Fillard, Nicholas Ayache, Pierre Clavelou, and Grégoire Malandain. SepINRIA : un outil d évaluation de la charge lésionnelle chez des patients atteints de sclérose en plaques. In Journées Internationales de la SFN : Sclérose en Plaques (Multiple Sclerosis), Paris, France, June Keyword(s): multiple sclerosis, AUTHOR = {Jean-Christophe Souplet and Christine Lebrun and Pierre Fillard and Nicholas Ayache and Pierre Clavelou and Grégoire Malandain}, BOOKTITLE = {Journées Internationales de la SFN : Sclérose en Plaques (Multiple Sclerosis)}, TITLE = {SepINRIA : un outil d évaluation de la charge lésionnelle chez des patients atteints de sclérose en plaques}, YEAR = {2007}, ADDRESS = {Paris, France}, MONTH = {June}, KEYWORDS = {multiple sclerosis, MedINRIA} } Automatic Lesion Segmentation algorithms Guillaume Dugas-Phocion, Miguel Angel Gonzalez Ballester, Christine Lebrun, Stà phane Chanalet, Caroline Bensa, Grà goire Malandain, and Nicholas Ayache. Hierarchical Segmentation of Multiple Sclerosis Lesions in Multi-Sequence MRI. In International Symposium on Biomedical Imaging: From Nano to Macro (ISBI 04), Arlington, VA, USA, April

39 @INPROCEEDINGS{dugas:isbi:2004, AUTHOR = {Guillaume Dugas-Phocion and González Ballester, Miguel Ángel and Christine Lebrun and Stéphane Chanalet and Caroline Bensa and Grégoire Malandain and Nicholas Ayache}, BOOKTITLE = {International Symposium on Biomedical Imaging: From Nano to Macro (ISBI 04)}, TITLE = {Hierarchical Segmentation of Multiple Sclerosis Lesions in Multi-Sequence MRI}, YEAR = {2004}, ADDRESS = {Arlington, VA, USA}, MONTH = April, ORGANIZATION = {IEEE}, KEYWORDS = {multiple sclerosis} } 38

40 Chapter 8 About Copyright (c) 2008 INRIA - ASCLEPIOS Research Team Project Management Jean-Christophe Souplet (Jean-Christophe.Souplet@sophia.inria.fr); Scientific Direction Christine Lebrun; Gregoire Malandain; Jean-Christophe Souplet; Developers Pierre Fillard; Jean-Christophe Souplet; Nicolas Toussaint; External Collaborators Dominique Aufauvre, Hospital Doctor, CHU Gabriel Montpied - Clermont-Ferrand, France; William Camu, Neurologist, CHU Gui de Ghauliac - Montpellier, France; Stephane Chanalet, Radiologist, CHU Pasteur - Nice, France; Pierre Clavelou, Neurologist, CHU Gabriel Montpied - Clermont-Ferrand, France; Mikael Cohen, House Physician, Nice Hospitals, Nice, France; Jean Gabrillargues, Radiologist, CHU Gabriel Montpied - Clermont-Ferrand, France; Christine Lebrun, Neurologist, CHU Pasteur - Nice, France; Note: The current release of SepINRIA is a beta version and should not be considered otherwise. Please report any problem with the software to Jean-Christophe.Souplet@sophia.inria.fr. A priori Atlases are part of SPM software ( and are included in SepINRIA with the authorization of John Ashburner of the Functional Imaging Laboratory of UCL. Redistribution and use in source and binary forms, with or without modification, is NOT ALLOWED without the authorization of the developers. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. SepINRIA 39

41 is a software for non-commercial use only. If you use this software for any purpose (e.g. publication), a reference to the program and the authors must be included ( THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER AND CONTRIBU- TORS AS IS AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FIT- NESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUD- ING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SER- VICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOW- EVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARIS- ING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 40