Bile duct segmentation from 3D CT image based on machine learning and probability map-assisted region growing

摘要

In this paper, we present our study on the bile duct segmentation from 3D CT volumes. In hepatobiliary surgery, it is required to know the spatial structure of the bile duct in advance. In our segmentation method, we introduce a region growing method assisted by probability map obtained from machine learning classification. At the first stage of our method, each voxel is classified as a voxel of the bile duct or not by the support vector machine. By using Platt's probabilistic outputs for support vector machines, we can acquire a probability map of the bile duct. At the second stage, we utilize the probability map to conduct a probability map-assisted region growing procedure to get the final segmentation result. In our experiments, the region growing procedure improved bile duct segmentation significantly (p=0.059). F-score increased from 0.55 to 0.58 by using the procedure.