Aorta Segmentation v2

OsiriX and Horos Course · Segmentation

In this lesson I demonstrate an automatic workflow for segmenting the aorta (and vascular structures) in OsiriX, to turn a “cluttered” dataset (with bones, organs and other structures) into a clean image, ideal for really nice 3D reconstructions (Volume Rendering). The idea is: select the contrast-filled aorta, expand that selection in 3D, fix the ROI (close holes and dilate) and then paint everything outside the ROI with negative density (like air). With that, the 3D becomes much easier to generate and to “sculpt”. Method overview Select the contrast-filled aorta using a 2D ROI Expand to 3D (up and down through the series) with Grow Region 3D Edit the ROI: Closing to close holes Dilatation to include a bit beyond the lumen (adventitia) Set Pixel Value (Outside ROI) to put negative density outside the selection Generate the 3D Volume Rendering and finish with quick cuts (scissors + Delete) Step by step (as in the video) 1) Prepare the view I'm on the axial image I adjust a vascular windowing (brightness/contrast) to make the selection easier 2) Create the automatic segmentation with an ROI I zoom in on a region where the aorta has good density (well opacified) I go to ROI → Grow Region 2D/3D Segmentation I choose 3D Grow Region (not the 2D one) Adjusting the range (HU) OsiriX selects based on a range (e.g., X HU units above and below the clicked point) I demonstrate by testing smaller values (e.g., 30) and larger ones In the example, 150 worked well to also capture branches with slightly lower density Then I have OsiriX compute the 3D selection and I can navigate up/down checking: celiac trunk superior mesenteric renal arteries selected along with the aorta. 3) Fix ROI defects (holes) and refine the selection Since “holes” appear inside the selection: I go to ROI → Brush → Closing I close holes of size ≤ 4 voxels Then, so as not to stay only in the lumen: I go to ROI → Brush → Dilatation I dilate by 2 voxels (in the example) to “fatten” the selection a bit 4) Paint everything outside the ROI with negative density Now comes the trick to “clean” the dataset: In ROI → Set Pixel Value I choose: Outside ROIs A negative value (e.g., -2048; -1024 would work too) This way, everything outside the ROI becomes “air” (negative density), making the 3D easier In the video I mention that the exam is large (about 2,600 images), so this calculation may take a while. 5) Generate the 3D and handle the “dynamic range” (large exam) After processing, I remove/hide the ROI to view the “clean” result I go to 3D Volume Rendering Since the exam is very large, a situation comes up where the scope/dynamic range becomes excessive I show two options: Resample (compresses the variation) Clip (cuts the extremes) — I choose Clip because it's faster for the computer Result: a segmented aorta with clearly visible branches and a much more “workable” 3D. Quick finishing (post-processing) Even with automatic segmentation, some things may remain: a bit of heart (due to contrast) vertebrae / pelvis Then I use: Scissors in the 3D select and press Delete Making cuts from different angles, you can “sculpt” quickly and reach a very clean result.