Inspiration
Manual identification of aortic branches is slow and difficult. We wanted a lightweight method that adapts to each patient without requiring a large training dataset.
What it does
Finding Artery detects contrast-filled vessels leaving the aorta. It reports each branch’s ostium, 5 mm seed point, initial direction, radius, and centerline, with an interactive 3D CT viewer for verification.
How we built it
We estimate blood intensity from the supplied aorta mask, grow locally through matching voxels, separate protrusions using connected components and watershed, then validate their length, shape, radius, direction, and leakage.
Challenges we ran into
Branches can touch, curve sharply, or remain close to the aorta. CT resolution and contrast also vary between patients, causing missed vessels, duplicate detections, and false growth into nearby structures.
Accomplishments that we're proud of
We built a CPU-efficient, patient-specific detector using physical measurements instead of fixed thresholds or machine learning. We also created an interactive viewer for examining predictions in three CT planes and 3D.
What we learned
Physical distances, 26-voxel connectivity, ostium contact geometry, and curved centerline length are essential. We also learned that validation must balance rejecting false positives without removing real oblique branches.
What's next for Finding Artery
We plan to test on more patients, improve difficult near-wall branches, add vesselness as supporting evidence, quantify accuracy against expert annotations, and connect detected seeds to full vascular segmentation.
Log in or sign up for Devpost to join the conversation.