MSF-Net: A Lightweight Multi-Scale Feature Fusion Network for Skin Lesion Segmentation

Area of Science:

• Medical image analysis
• Computer vision
• Dermatology

Background:

• Accurate skin lesion segmentation is crucial for early melanoma diagnosis.
• Challenges include variations in scale, shape, contrast, and boundary definition in dermatological images.

Purpose of the Study:

• To propose a Multi-scale Feature Fusion Network (MSF-Net) for improved skin lesion segmentation.
• To address the limitations of existing methods in handling complex lesion characteristics.

Main Methods:

• Developed MSF-Net based on a Comprehensive Attention Convolutional Neural Network (CA-Net).
• Incorporated spatial attention mechanisms within residual connections to focus on critical image regions.
• Utilized Multi-scale Dilated Convolution Modules (MDC) and Multi-scale Feature Fusion Modules (MFF) for enhanced contextual information extraction and adaptive receptive field adjustment.

Main Results:

• MSF-Net demonstrated high validity on the ISIC2018 dataset.
• Ablation studies confirmed the effectiveness of the proposed attention, MDC, and MFF modules.
• Comparative experiments showed MSF-Net achieves superior segmentation performance with fewer parameters compared to existing advanced networks.

Conclusions:

• MSF-Net offers an effective solution for challenging skin lesion segmentation tasks.
• The proposed network architecture improves diagnostic accuracy for melanoma through enhanced image analysis.
• MSF-Net presents a computationally efficient approach for dermatological image segmentation.