Coronary Artery Vessel Tree Segmentation Using Transfer Learning from CT Angiography Images

Automatic segmentation of coronary arteries is crucial for precise diagnosis and treatment planning in cardiovascular imaging. In this study, we employ a deep learning framework based on the nn-UNet framework for coronary artery segmentation. Our approach leverages annotated datasets, namely ASOCA and ImageCAS, to enhance segmentation accuracy using transfer learning. Additionally, we apply post-processing techniques to further refine the segmentation results. Evaluation on both datasets demonstrates improvements in segmentation accuracy, highlighting the effectiveness of our method in handling complex anatomical structures such as coronary arteries. Notably, the incorporation of transfer learning led to a significant enhancement in segmentation performance, underscoring its value in coronary arteries segmentation. The proposed approach leveraging from transfer learning and post-processing achieves Dice score of 0.856 and a Hausorf distance of 14.5 on the ASOCA dataset.