MFA-Net: Multiple Feature Association Network for medical image segmentation

Area of Science:

• Medical imaging analysis
• Computer-aided diagnosis
• Deep learning applications

Background:

• Accurate medical image segmentation is vital for computer-aided diagnosis but challenging due to image variability.
• Existing deep learning models struggle with capturing diverse features for precise segmentation.

Purpose of the Study:

• To develop a novel deep learning network, the Multiple Feature Association Network (MFA-Net), for enhanced medical image segmentation.
• To improve the accuracy of segmentation for various medical imaging tasks.

Main Methods:

• Implemented an encoder-decoder architecture with skip connections.
• Integrated a Parallelly Dilated Convolutions Arrangement (PDCA) module for feature extraction.
• Incorporated a Multi-scale Feature Restructuring Module (MFRM) for feature fusion.
• Utilized Global Attention Stacking (GAS) modules to enhance global perception.

Main Results:

• MFA-Net demonstrated superior performance across four diverse medical image segmentation tasks: intestinal polyp lesions, liver tumors, prostate cancer, and skin lesions.
• Experimental results confirmed MFA-Net's effectiveness in both global positioning and local edge recognition compared to state-of-the-art methods.
• Ablation studies validated the contribution of individual modules to the overall performance.

Conclusions:

• The proposed MFA-Net significantly advances medical image segmentation accuracy.
• MFA-Net's novel architecture, incorporating PDCA, MFRM, and GAS modules, effectively addresses challenges in medical image variability.
• This network shows great potential for improving computer-aided diagnosis systems.