Local edge-enhanced active contour for accurate skin lesion border detection

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision

Background:

• Dermoscopy is crucial for skin cancer diagnosis, particularly for pigmented lesions.
• Accurate skin lesion border detection is essential for analyzing dermoscopic features.
• Manual border delineation by dermatologists is subjective, leading to variations and inefficiency.

Purpose of the Study:

• To develop an automated method for skin lesion border detection in dermoscopic images.
• To overcome the limitations of manual segmentation, including inter-observer variability and tediousness.

Main Methods:

• A novel skin lesion border detection method utilizing a robust edge indicator function.
• Implementation based on a meshless method, specifically employing smoothed particle hydrodynamic (SPH) kernels.
• Integration of a probability map to prevent contour leakage during segmentation.

Main Results:

• The proposed algorithm outperforms existing state-of-the-art image segmentation methods.
• Achieved superior boundary generation compared to prominent methods, with a Dice score of 0.886 ±0.094 and Jaccard score of 0.807 ±0.133 against dermatologist-drawn ground truth.

Conclusions:

• Smoothed particle hydrodynamic (SPH) kernels are effective as edge features for active contours segmentation.
• Probability maps enhance segmentation accuracy by preventing contour leakage into the region of interest.