Sites@Duke | sites.duke.edu



PROTOCOLSemi-automated tumor segmentation with the “GrowCut” Tool in 3D Slicer[BlockerSJ 2018, stephanie.blocker@duke.edu ]General Description:This protocol describes a basic method for semi-automated tumor volume definition of DICOM image files using 3D Slicer. This means that in defining tumors in the image, user input is still required but is limited compared to other methods (such as hand-drawn, slice-by-slice ROI definition).Notes:The following method requires initial input from the user, but removes any further user decisions after utilization of the “GrowCut” tool. This does not limit the user from making changes after use of the tool. However, caution should be used when applying post-tool modifications to volumes of interest, as this can incorporate a fluctuating bias on a per tumor basis.To reduce post-tool correction necessity, this protocol utilizes a second set of tools to deal with “islands”. Islands are considered areas that were highlighted in the ROI which are unlikely to be true. This is calculated by the software, and helps remove discontinuous, disconnected, and diffuse portions of the volume which can result from the “Grow/Cut” tool. Although this tool is not absolute, it is simple to incorporate and helps reduce user bias.This protocol contains the following sections:Importation of DICOM data into 3D SlicerApplying a bias correction (for use with images generated with MR surface coils)Casting Scalar VolumeIdentifying tumor and non-tumor regions in three orientationsApplying the “GrowCut” toolApplying the “Remove Islands” tool setMeasuring tumor volumeGenerating a 3D model (optional)TroubleshootingAdditional help navigating 3D SlicerViewing navigation and linking of orthogonal viewsSaving a variety of file typesScreenshots and scene snapshotsGENERAL PROTOCOLImportation of DICOM data into 3D SlicerUpon opening of the 3D Slicer software, the user finds the typical viewing platform, along with options in the Module Panel (left side) for accessing data:To import DICOM files, select “Load DICOM data” from the Module panel. Alternatively, select the DCM button at the top left of the screen: The user can access previously loaded DICOMs from the resulting DICOM browser. Once the desired study is selected, choose “Load” to open. Once a DICOM file has been imported into the software, it will be available to open via the DICOM browser.To import new DICOM files, select “Import” in the upper left corner of the DICOM browser. This will prompt the user to select a directory from their files. DICOMs are typically saved as a folder (“directory”) which contains the entire stack of images. To import, locate and select the entire folder to import. This should import the stack of slices as a 3D volume that is now available in the DICOM browser.Once the image is loaded, the user can scroll through the orthogonal views, as well as adjust contrast by holding the right click down, and dragging across the viewscreen.Note: It is recommended at this point that the user link the 3 orthogonal views. To do this, select the pushpin at the top of one of the viewing panels to pin the slice controls to the panel:To link all views, select the “link/unlink slice controls” button to activate link: Casting Scalar VolumeIn order to perform “GrowCut” in Slicer, the image needs to be filtered to a “short type” image. If this step is not performed, the subsequent steps of this protocol will fail.From the Module search, under “All Modules” find and select the “Cast Scalar Volume” module:For the Input Volume, select the original image.Under Output Volume, select “Create new volume as…” and enter a name for the new, “short-filtered” version.Select “Short” as the output type, and select “Apply”. This process will take a moment. Depending on the image, the contrast may appear to change, resulting in a black or washed out image. The image will still be intact – simply adjust the contrast of the short-filtered image to return it to its original appearance.Applying a bias correction (for use with images generated with MR surface coils)In studies which utilize a surface coil for MR, images may have an inherent bias throughout the sample, seen as a reduction in signal as distance from the coil increases (see example of a phantom shown below). In these cases, a bias correction can be applied to the image in Slicer.Under the Module search tab, locate and select the “N4ITK MRI Bias correction” module:In the N4ITK module, select the newly short-filtered volume as the input image. Create a new output volume by selecting “Create new volume as…” and provide a new name for the bias corrected volume. If a labelmap or volume representing the applied bias correction is desired, create a new file by selecting “Create new labelmap as…” or “Create new volume as…” and rename the file appropriately. (See image below for reference).Enter the bias correction parameters which apply to the system being used. At the CIVM, when using the Bruker 7T magnet, the bias correction is calculated using 1000,1000 iterations and a 0.0002 convergence threshold. (See image below for reference).When parameters are entered, select “Apply”. This will take a moment to complete.When the bias correction is complete, the user can compare the bias-corrected image to the original in the slice control panel. Select the double arrow on the left to show the view options.The middle and bottom selections represent the foreground and background selections, respectively. Select the bias corrected volume as the foreground, and the original image as the background.Using the slide bar on the left side of the slice control panel, the user can toggle between the images to view the effect of the bias correction.If the bias correction is satisfactory, save the related volumes by selecting “Save” at the top left side of the screen.In the Save window, select all of the files to keep, including the new bias-corrected file. The location of the saved files can be designated by selecting the “Change directory for saved files” button at the bottom of the prompt.Save the corrected file as a .nrrd for future use in Slicer. If needed, additional copies of the file can be saved as other file types by repeating the save process and selecting a different type from the dropdown option.From this point, if performing semi-automated volume analyses, the user should continue to Section IV, using the bias-corrected version of the MR image.Remember, semi-automated volume detection operates by detecting boundaries and changes in signal intensity. Thus, an inherent bias in the image will affect the ability of the software to generate ROIs around a volume which reasonably spans a bias. Thus, the appropriate bias correction should be applied to such images.Identifying tumor and non-tumor regions in three orientationsTumor/ROI identification in Slicer is performed by creating an image label map. To begin, enter the Editor Module through the search function, or by pressing the pen icon:Entry into the Editor module will prompt the selection of a color map. This is strictly preference. This protocol will proceed with the default option “Generic Anatomy Colors”.Ensure that the proper volumes are selected, with the Master Volume representing the working image, and the “Merge Volume” representing the label of that image (“-label”). At any time, a new label map (such as for a new ROI) can be generated by selecting “Create new Labelmap volume” as a new Merge Volume. Note: Each volume that is created will appear separately when saving (be sure to check all desired label maps when saving!).To begin identifying tumor regions, select the paintbrush tool under “Edit Selected Label Map”. The tool size/radius and shape can be modified at the bottom of the module.Locate the tumor in all 3 orthogonal views. To coordinate each view pane around a centralized location, or suspected tumor, hover the mouse over that region in one view while holding down the shift key. This should center this area in all three views. Continue to hold shift and hover to move in all three views as you hover and move the mouse.Once the tumor is located, define the following regions:DEFINING THE TUMORIn the label map module, change the paintbrush color to reflect tissue designation. In this case, select “9 – foreign object” to delineate tumor:In one of the three views, use the paintbrush tool to paint INSIDE the boundaries of the suspected tumor. The software will register these pixels as those of interest, so it is important to NOT deviate into extratumoral space. Instead, leave a safe margin at the edge of the tumor to ensure that extraneous pixels are not included in subsequent steps.Repeat this process in all three orthogonal views, while taking care to remain within the tumor volume in each view.If a mistake is made in drawing, actions can be undone/redone with buttons available on the module:DEFINING NON-TUMOR TISSUEOnce the tumor had been identified in all three views, the software must also be told which areas represent tissues to exclude. This is accomplished by defining “waste” regions.Change the paintbrush designation to “10 – waste”.As before, define areas of waste in all three orthogonal views, as they relate to the tumor. To do this, carefully identify regions surrounding the tumor which are definitely not tumor. It is best to surround the tumor on all sides, taking care to NOT include any pixels that may be tumor. Thus, between both colors, there will be an open margin that will represent the tumor margin.Repeat this process in all three orthogonal views.Applying the “GrowCut” toolOnce the user feels confident in the 3D identification of tumor and non-tumor tissues, they can begin automated propagation of the ROI.Select the “GrowCut” tool, and reselect the yellow (“9 – foreign object”) color. Choose “Apply”. This process will take a moment.When complete, the area within the tumor, as well as extratumoral area will be propagated in their respective colors.The propagated “tumor” VOI will likely have extraneous and disconnected regions that appear throughout the waste regions. If left, these satellite “islands” will be factored into the volume measurements.Although the user could manually erase islands and clean edges, Slicer provides a tool set to help with these issues. This removes additional bias on the part of the user.Applying the “Remove Islands” tool setTo remove islands, and create more cohesive tumor edges (remove scatter at ROI borders), apply the “Remove Islands” tools, under Edit Selected Label Map:This includes two tools:First, select for connectivity by pressing “Apply Connectivity Method”. This removes the designated waste regions, and begins removing disconnected islands.Next, select “Apply Morphology Method”. This tool helps remove satellite islands, and also helps correct “fuzzy edges” by assuming a more compartmentalized morphology.NOTE: This tool takes a few moments to operate. Importantly, the software often appears to finish the operation with no changes. This is simply a delay – waiting about 30 seconds will reveal the actual output. The user will be able to notice a difference, before and after. If no changes are noticed, the operation has not finished – do not interrupt prematurely!When completed, save the label map. Measuring tumor volumeOnce the label map is generated, Slicer is able to calculate the volume of the tumor region.Under All Modules, locate and select the “Label Statistics” Module:In the module, select the working image as the Grayscale volume. Select the label map generated in the previous step as the Label Map.Select Apply. This will generate a table which provides the measurements of the volume of interest.Generating a 3D model (optional)If desired, the user can generate a 3D model of the selected ROI generated in the previous steps.In the Editor Module, under the Edit Selected Label Map, choose the “Make Model” effect:Ensure that the appropriate label is selected, and hit apply. This will generate a 3D model of the VOI which appears in the 3D viewing panel. To see the 3D model in the context of the orthogonal views, select the “Open eye” icon in the slice controls:TroubleshootingAdditional help navigating 3D SlicerViewing navigation and linking of orthogonal viewsSaving a variety of file typesScreenshots and scene snapshots ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download