Chongqing Medical University (Zeng et al.)
Self-supervised model that performs single-frame digital-subtraction-angiography-style vessel/background separation directly from a single live (non-subtracted) coronary angiogram frame, then supports fine-tuned coronary vessel segmentation. A U-Net-style network is pretrained via an image-to-image translation objective on 58,128 unannotated angiography DICOM series (3,756 patients), then fine-tuned for vessel segmentation on just 40 expert-annotated frames, reaching a Dice of 0.828 on the held-out fine-tuning set and a new state-of-the-art Dice of 0.755 on the public XCAD benchmark. Intended to help clinicians visualize potential stenosis sites without requiring true two-frame digital subtraction acquisition.
Architecture
CNN (2D)
Self-supervised U-Net trained via an image-to-image translation objective for single-frame vessel/background subtraction, subsequently fine-tuned for supervised coronary vessel segmentation
Framework
PyTorch
Added to catalog
2026-08-12
40 coronary angiography images with fine-grained, expert-annotated vessel segmentation masks, used to fine-tune DeepSA for supervised vessel segmentation.
58,128 unannotated coronary angiography DICOM series from 3,756 patients at the Second Affiliated Hospital of Chongqing Medical University, used for self-supervised pretraining of single-frame vessel/background subtraction.
XCAD (public X-ray coronary angiography vessel-segmentation benchmark)
Public benchmark of X-ray coronary angiography images with vessel-segmentation ground truth, used by DeepSA as an external, out-of-distribution evaluation set.
Single-frame digital-subtraction-angiography-style image synthesized directly from one live (non-subtracted) coronary angiogram frame
Binary segmentation of the coronary vessel tree in X-ray coronary angiography frames