Automated abdominal artery branch labeling via topological hypergarph neural networks

Accurate anatomical labeling of abdominal arteries plays a pivotal role in diagnosing abdominal vascular diseases. But existing methods have certain limitations, such as insufficient focus on global correlations and an over-reliance on prior knowledge. In this study, multiple branches of the abdominal aorta are grouped into a single hyperedge, and a hypergraph neural network (HGNN) is employed to extract branch geometric features, spatial features, and topological encoding features obtained through dual branch traversals. Experiments are conducted on the public dataset TotalSegmentator. The experimental results show that our method can achieve the best results on the main branch Ao and branches at all levels.