CF-VCENet: Coarse-to-Fine Vascular Connectivity Enhancement Network for Hepatic Vessel Segmentation in MR Images

Motivation: Vessel location must be pinpointed for precise avoidance during probe insertion for liver ablation. Goal(s): Our goal was to segment the hepatic vessel from MR images and to ensure the connectivity of the segmentation results. Approach: Hepatic vessel MR images were obtained from the records of 105 patients. Coarse-to-fine vascular connectivity enhancement algorithm was trained and tested using a five-fold cross-validation method. Results: Results demonstrated that our two-stage algorithm improved the connectivity of vessel segmentation results, with the dice coefficient increasing by up to 1.8% compared to the initial segmentation. Impact: Accurate segmentation results and enhanced connectivity provide a basis for hepatic vessel location and modeling. The results can assist doctors in preoperative planning while reducing the risk of damage to normal tissue in patients.