An updated arXiv preprint describes a self-supervised pre-training method designed for X-ray angiogram analysis. The authors note that X-ray angiography remains the gold standard for imaging cardiovascular disease, but deep learning models in this area are held back by a severe shortage of annotated data.
The proposed solution is vascular anatomy-aware pre-training, which aims to make use of structural information from the vessels themselves. By learning from unlabeled or weakly labeled angiograms in this way, the method could reduce the need for large manually annotated datasets.
Because the abstract is the only available source, the article does not compare this approach with other pre-training strategies. The stated motivation is clear, however: better use of vascular anatomy may help deep learning models perform well even when labeled angiograms are scarce.