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

The article addresses the application of the SAM2 model (developed by Meta in 2024) for segmenting coronary arteries in X-ray fluoroscopic images. Direct application of the model is challenging due to specific medical imaging difficulties, including noise from patient movement, the projection-based nature of X-ray imaging, and low contrast between vessels and background. The research team performed fine-tuning of the MedSAM2 model on annotated coronary angiograms. On the ARCADE validation set (200 images), the fine-tuned model achieved a Dice score of 0.767 compared to 0.033 with zero-shot learning. When tested on 10 fluoroscopic video studies from the CoronaryDominance database, the model coherently tracked vessels and avoided false segmentation of ribs, stents, and bypass grafts in 9 out of 10 cases. The code and fine-tuned model are available on public platforms.