Liver vessel segmentation based on inter-scale V-Net.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Deep Learning

Background:

• Accurate segmentation of liver vessels is crucial for preoperative planning and diagnosing liver diseases.
• The irregular structure of liver vessels presents a significant challenge for precise segmentation.
• Existing methods often struggle with preserving spatial details and integrating multi-scale information.

Purpose of the Study:

• To develop an improved V-Net network for accurate liver vessel segmentation.
• To enhance the network's ability to capture fine details and semantic features.
• To improve the overall performance of liver vessel segmentation from CT images.

Main Methods:

• An improved V-Net architecture incorporating dilated convolution to enlarge the receptive field without losing spatial information.
• Integration of a 3D deep supervision mechanism to accelerate network convergence and improve semantic feature learning.
• Implementation of inter-scale dense connections in the decoder to preserve high-level semantic information and fuse multi-scale features.
• Validation using the public 3Dircadb dataset for liver vessel segmentation experiments.

Main Results:

• The proposed improved V-Net method achieved an average Dice score of 71.6% and a sensitivity of 75.4%.
• These results demonstrate superior performance compared to the original V-Net network.
• The method successfully segmented both labeled and unlabeled liver vessels from CT images with high accuracy.

Conclusions:

• The improved V-Net network offers an effective solution for automated and accurate liver vessel segmentation.
• This advancement holds significant potential for improving preoperative planning and computer-aided diagnosis of liver conditions.
• The proposed network architecture enhances the segmentation of complex vascular structures in medical imaging.