Inspiration Aortic branch anatomy varies a lot between patients, and surgeons repairing the aorta need to know exactly where each artery leaves it. We wanted to see how far classical image processing (no GPU, no training data) could get toward automating that discovery directly from CT scans.
What it does Given a CT scan and a mask of just the parent aorta, our tool automatically detects every artery branching directly off it without being told in advance how many there are or where. For each one it finds, it outputs the origin point, initial direction, and vessel radius, all in real-world physical coordinates.
How we built it The pipeline runs in stages: first it scans the aorta's outer wall for spots that stay at blood-vessel brightness rather than fading into surrounding tissue, then it walks outward from each candidate to check it actually behaves like a vessel rather than noise, then separates confirmed candidates into individual branch regions and traces each one's centerline to measure it. Everything is built with SimpleITK, NumPy, and SciPy, so it runs on a standard laptop with no GPU.
Challenges we ran into Contrast brightness varies a lot case to case, the same intensity threshold that correctly finds a branch in one scan can either miss a dimmer one or pick up false positives from calcified plaque in another. Tuning that balance without overfitting to a single example was the hardest part.
What's next Replacing the hand-tuned brightness/shape thresholds with values calibrated against labeled reference cases, and extending validation across more of the dataset.
Log in or sign up for Devpost to join the conversation.