Automatic 3D Segmentation of Liver Blood Vessels Using Deep Learning

Abstract

This study addresses the time-consuming and delicate nature of manually and automatically dividing hepatic blood vessels. The objective is to develop a rapid, highly accurate, and efficient autonomous division using deep learning. The proposed DeepVesselNet-FCN architecture involves data collection, pre-processing, cross-filter creation, extreme class balancing, and network execution. This model achieves a Dice coefficient of 0.8549, demonstrating good accuracy. This study concludes by offering a practical application, a simple program enabling physicians to automatically and accurately divide liver blood vessels in three dimensions.