JAAL-Net: a joint attention and adversarial learning network for skin lesion segmentation

Area of Science:

• Medical image analysis
• Artificial intelligence in dermatology

Background:

• Accurate skin lesion segmentation is crucial for melanoma diagnosis and treatment.
• Current methods struggle to differentiate lesions from artifacts like hairs and blood vessels, impacting performance.

Purpose of the Study:

• To develop a novel network, JAAL-Net, for enhanced skin lesion segmentation.
• To improve the accuracy and reliability of automated melanoma detection systems.

Main Methods:

• Proposed JAAL-Net integrates a generator (LF-Net) with an encoder-decoder structure and attention modules.
• The network utilizes convolutional block attention and contour attention for lesion and boundary information.
• A discriminant dual attention network enhances prediction confidence.

Main Results:

• JAAL-Net achieved high Intersection over Union (IoU) scores: 90.27% (ISBI2016), 89.56% (ISBI2017), and 80.76% (ISIC2018).
• The model effectively captured detailed lesion and boundary information.
• Demonstrated improved segmentation accuracy and prediction confidence.

Conclusions:

• JAAL-Net significantly enhances skin lesion segmentation performance.
• The approach provides a valuable tool to assist physicians in accurate melanoma diagnosis.