TPMA-Net: Topology-Preserving Multi-Scale Aggregation Network for Liver Segments Based Vascular Territory

The liver segments are crucial foundation for liver transplantation surgery, and the vascular territory-based segments is a novel theory proposed in the medical field in recent years. This method fully considers the impact of oxygen supply by vascular segments on the liver when segments the liver, which significantly aids in improving the postoperative quality of life for patients. Due to the difficulty in establishing a spatial mapping relationship between liver CT image vascular voxels and segment labels, using traditional methods for direct segments yields unsatisfactory results. Currently, deep learning segments algorithms based on vascular territories are almost semi-automatic. To address these issues, this paper proposes a novel automated liver segments method based on Topology-Preserving Multi-Scale Aggregation (TPMA). This method first obtains the skeleton of the vascular segments to construct a topological structure, and then uses TPMA-Net to achieve segments of the portal vein centerline. Finally, the liver is divided into liver segments using a three-dimensional minimum distance method. The results of experiments show that the final segments accuracy reached 93.9%, and our proposed method has initially achieved the requirements of automatic segments of the liver based on vascular territories.