Skin Lesion Segmentation Using an Ensemble of Different Image Processing Methods

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision

Background:

• Skin cancer cases are rising, making early detection critical.
• Dermoscopic images contain artefacts (hair, markers, poor boundaries) that hinder automated analysis.
• Existing segmentation methods struggle with diverse skin lesion artefacts.

Purpose of the Study:

• To develop an accurate and efficient automated method for skin lesion segmentation.
• To overcome limitations of single thresholding methods in handling image artefacts.
• To improve early detection of skin diseases through precise lesion delineation.

Main Methods:

• Proposed an ensemble-based method for skin lesion segmentation.
• Optimized threshold selection using an objective function.
• Integrated multiple state-of-the-art thresholding algorithms (Otsu, Kapur, Harris hawk, grey level).

Main Results:

• The proposed ensemble method achieved a superior dice score of 0.89 (p ≤ 0.05).
• Outperformed individual methods: Otsu (0.79), Kapur (0.80), Harris hawk (0.60), grey level (0.69), and active contour model (0.72).
• Demonstrated effectiveness on the ISIC 2016 dataset.

Conclusions:

• Ensemble-based segmentation effectively addresses artefacts in dermoscopic images.
• The proposed method offers a significant improvement over existing techniques for skin lesion analysis.
• Accurate segmentation is vital for advancing automated skin disease diagnosis and early detection.