DMCGNet: A Novel Network for Medical Image Segmentation With Dense Self-Mimic and Channel Grouping Mechanism.

Area of Science:

• Medical Image Analysis
• Computer Vision
• Artificial Intelligence

Background:

• Deep learning dominates Medical Image Segmentation (MIS) for automatic feature extraction.
• Current methods struggle with plain network designs and diverse medical image targets, limiting semantic feature extraction.

Purpose of the Study:

• To propose a novel Dense Self-Mimic and Channel Grouping based Network (DMCGNet) for enhanced Medical Image Segmentation.
• To improve the extraction and utilization of semantic features in medical images.

Main Methods:

• Introduced the Pyramid Target-aware Dense Self Mimic (PTDSM) module for deeper feature representation without increasing parameters.
• Developed the Channel Split based Feature Fusion Module (CSFFM) for efficient feature reuse and multi-scale target adaptation using channel grouping.
• Integrated Deep Supervision with Group Ensemble Learning (DSGEL) for adequate network training.

Main Results:

• The proposed DMCGNet demonstrated state-of-the-art performance across four medical image segmentation datasets.
• PTDSM module effectively enhanced feature representation.
• CSFFM module improved multi-scale target adaptation and feature reuse.

Conclusions:

• DMCGNet offers a significant advancement in Medical Image Segmentation accuracy and efficiency.
• The novel modules (PTDSM, CSFFM) and training strategy (DSGEL) contribute to superior performance.
• The model shows strong potential for assisting clinicians by reducing labor and standardizing segmentation outcomes.