A hybrid approach for accurate skin lesion segmentation using LEDNet and Swin-UMamba

Area of Science:

• Medical Imaging
• Computer Vision
• Dermatology

Background:

• Accurate skin lesion segmentation is crucial for effective skin cancer detection.
• Existing segmentation methods face challenges with irregular boundaries, textures, and artifacts in skin lesions.

Purpose of the Study:

• To propose a novel hybrid model for multiscale skin lesion segmentation.
• To improve the accuracy and robustness of skin lesion delineation by integrating edge-accurate LEDNet and Swin-UMamba.

Main Methods:

• Developed a hybrid model integrating LEDNet for edge accuracy and Swin-UMamba (Mamba-based encoder with VSS block) for multiscale feature extraction.
• Evaluated the model on the Ph[Formula: see text], ISIC-2017, and ISIC-2018 skin cancer datasets.

Main Results:

• Achieved high performance across datasets: ISIC-2017 (DSC: 0.9734, Sensitivity: 0.9697, Specificity: 0.9858, Accuracy: 0.9847), ISIC-2018 (DSC: 0.9753, Sensitivity: 0.9494, Specificity: 0.9902, Accuracy: 0.9713), and Ph (DSC: 0.9801, Sensitivity: 0.9892, Specificity: 0.9966, Accuracy: 0.9932).
• The hybrid model demonstrated superior capability in capturing complex lesion boundaries and textures compared to standalone methods.

Conclusions:

• The proposed hybrid framework shows significant potential for advancing skin lesion segmentation.
• This method is promising for clinical dermatology, offering improved diagnostic capabilities through accurate image analysis.