Deep learning based coronary vessels segmentation in X-ray angiography using temporal information

Haorui He, Abhirup Banerjee, Robin P. Choudhury, Vicente Grau

Abstract: Highlights•We propose TVS-Net, a 2D+time framework for vessels segmentation from ICA.•We use a connectivity-preserving loss to ensure vascular integrity and accuracy.•TVS-Net outperforms benchmarks with 83.4% Dice and 84.3% recall on 323 ICA images.•TVS-Net outperforms manual annotations, achieving 86.3% recall.•TVS-Net outperforms benchmarks by 4% in AUPRC on 60 out-of-distribution ICAs.