Liver Semantic Segmentation Method Based on Multi-Channel Feature Extraction and Cross Fusion.

Area of Science:

• Medical Image Analysis
• Computer Vision
• Artificial Intelligence

Background:

• Accurate liver segmentation is crucial for diagnosing and planning treatment for liver diseases.
• Current segmentation methods struggle with the liver's complex anatomy and patient variability, limiting feature extraction and fusion.
• Challenges in precise liver segmentation hinder effective clinical applications.

Purpose of the Study:

• To develop an improved U-Net-based semantic segmentation method for enhanced liver segmentation.
• To address limitations in feature extraction and fusion in existing liver segmentation techniques.
• To improve the accuracy and clinical utility of automated liver image analysis.

Main Methods:

• Implemented a multi-scale input strategy and a multi-scale convolutional attention (MSCA) mechanism in the encoder for improved feature representation.
• Integrated an atrous spatial pyramid pooling (ASPP) module in the bottleneck for capturing multi-receptive field features and global pooling for contextual information.
• Utilized a Channel Transformer module to replace traditional skip connections, enhancing feature interaction and reducing the semantic gap between encoder and decoder.

Main Results:

• The proposed method achieved a high Intersection over Union (IoU) of 0.9315 on integrated public datasets for liver segmentation.
• Demonstrated superior performance compared to other mainstream liver segmentation approaches.
• Validated the effectiveness of the optimized feature extraction and fusion mechanisms.

Conclusions:

• The improved U-Net-based method offers a novel and effective solution for precise liver semantic segmentation.
• The enhanced feature extraction and fusion strategies significantly boost segmentation accuracy.
• This approach holds substantial clinical value for the diagnosis and treatment planning of liver diseases.