EAAC-Net: An Efficient Adaptive Attention and Convolution Fusion Network for Skin Lesion Segmentation

Area of Science:

• Medical Image Analysis
• Computer Vision
• Dermatology

Background:

• Accurate skin lesion segmentation is crucial for melanoma diagnosis and quantitative analysis.
• Existing methods struggle with local-global feature extraction, challenging lesions, and computational complexity.

Purpose of the Study:

• To propose an efficient adaptive attention and convolutional fusion network (EAAC-Net) for improved skin lesion segmentation.
• To address limitations in feature extraction, handling difficult cases, and computational efficiency.

Main Methods:

• Developed EAAC-Net with parallel encoders: Efficient Adaptive Attention Feature Extraction Module (EAAM) for global dependencies and Efficient Multiscale Attention-based Convolution Module (EMA·C) for local features.
• Incorporated a Reverse Attention Feature Fusion Module (RAFM) to refine boundary information.
• Validated on ISIC 2016, ISIC 2018, and PH² datasets.

Main Results:

• EAAC-Net demonstrated superior segmentation performance compared to existing methods.
• The network effectively extracts local features with global information and handles challenging lesions.
• Reduced computational complexity through adaptive token filtering in EAAM.

Conclusions:

• EAAC-Net offers a significant advancement in skin lesion segmentation accuracy and efficiency.
• The proposed architecture effectively integrates attention mechanisms and convolutional features.
• EAAC-Net provides a promising tool for melanoma analysis and computer-aided diagnosis.