Automatic Aorta Segmentation

OsiriX and Horos Course · Segmentation

In this lesson I show how to do automatic segmentation of the aorta (and other structures) in OsiriX, to turn a “cluttered” dataset into a clean image, ready for really nice 3D reconstructions (Volume Rendering). The logic of the method is: Select a point with good contrast and create an ROI that grows in 3D (up and down through the series) Edit the ROI (close holes and dilate) to improve the selection Turn everything outside the ROI into negative density (like air), making the 3D easier Generate the 3D Volume Rendering and finish by removing whatever is left (scissors + Delete) 1) Create the 3D segmentation of the aorta (Grow Region) I set a vascular windowing I zoom in on a point where the aorta is well opacified I go to ROI → Grow Region 2D/3D Segmentation I choose 3D Grow Region Adjusting the range (HU) OsiriX selects a range in HU above/below the clicked point I test smaller and larger values to understand the effect In the example, 150 works well to select the aorta and branches with slightly lower density Then I compute the selection and check, navigating up/down, that the following were included: celiac trunk superior mesenteric renal arteries 2) Refine the ROI: close holes and dilate Since holes appear inside the selection: ROI → Brush → Closing I close holes of size ≤ 4 voxels So it is not limited to the lumen: ROI → Brush → Dilatation I dilate by 2 voxels to include a bit beyond the contrast (adventitia) 3) Paint everything outside the ROI with negative density ROI → Set Pixel Value I select Outside ROIs I apply a negative value (in the video: -2048; it could be -1024) This way, everything outside the aorta becomes “air”, which makes the 3D much simpler to build. (In the example it is a large study, with ~2600 images, so the computation may take a while.) 4) Generate the 3D and deal with a large study (Clip vs Resample) I go to 3D Volume Rendering Since the dataset is large, a behavior related to the dynamic range appears I show that you can: Resample (compress the variation) or Clip (cut the extremes) — I choose Clip because it is faster Result: a segmented aorta with clearly visible branches and a “clean” 3D image to work with. 5) Quick finishing: scissors + Delete If there is still something left: vertebrae/pelvis heart (it can pick up contrast too) I finish with: scissors tool in the 3D view select and press Delete It is a quick post-processing step and the result is quite impressive.