Fine-Tuning SAM2 for Coronary Artery Segmentation in X-Ray Fluoroscopy

SAM2 [Ravi et al., 2024] (Meta, 2024) provides a strong starting point for segmentation, but given the unique challenges in medical imaging (noise from patient movement, the projection-based nature of X-ray fluoroscopy, and low contrast between vessels and background), direct application is difficult. We fine-tune MedSAM2 [Ma et al., 2025] on annotated coronary angiograms and apply it to video data for point-of-care use. On the ARCADE validation set [Popielarski et al., 2024] (200 images), the fine-tuned model achieves Dice 0.767 ± 0.082 compared to 0.033 zero-shot. In 10 fluoroscopic video studies from CoronaryDominance [Danilov et al., 2025], it tracks vessels coherently and avoids false segmenting of ribs, stents, and bypass grafts in 9 of 10 studies. Code is available here and the fine-tuned checkpoint here .