BDU-Net: A New Application of U-Net to the Segmentation of Bile Ducts from Cholangio-MRI Images

This article introduces a novel approach to bile duct automatic segmentation from cholangio-MRI images. 3D reconstruction of bile ducts could be of great help for critical therapeutic interventions, especially for endoscopic retrograde cholangiopancreatography (ERCP). The approach is based on employing the U-Net neural network, incorporating a new loss function (BDU-Net). The study, conducted on 24 patient cases, utilized quantitative metrics and qualitative assessment to evaluate the segmentation results obtained by the proposed method against ground truth. Experimental results demonstrated that our model outperforms existing segmentation algorithms such as V-Net and U-Net++, achieving a Dice score of 0.72, representing a 16% improvement compared to the advanced U- Net++ segmentation method. From a qualitative point of view, our BDU-Net method generates fewer than one false communicating intrahepatic ducts (FC IHDs) and fewer than one false non-communicating intrahepatic ducts (FnC IHDs).