Preprocessing Effects on Performance of Skin Lesion Saliency Segmentation

Area of Science:

• Dermatology and Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Melanoma remains a challenging skin cancer despite advances in immunotherapy.
• Accurate melanoma diagnosis relies on effective segmentation of skin lesions in dermoscopic images.
• Existing segmentation methods struggle with the heterogeneous properties of dermoscopic images, often requiring complex preprocessing.

Purpose of the Study:

• To investigate the impact of image preprocessing on a saliency-based skin lesion segmentation method.
• To develop and validate a segmentation technique that eliminates the need for preprocessing in dermoscopic images.

Main Methods:

• A novel segmentation method combining color histogram clustering for initial region homogeneity and saliency map computation (integrating color contrast, ratio, spatial features, and central prior).
• Otsu thresholding and morphological analysis were used for artifact removal.
• The method was evaluated on benchmarking datasets, comparing performance with and without preprocessing.

Main Results:

• The proposed method successfully segmented skin lesions in dermoscopic images without requiring preprocessing.
• Performance was competitive with leading supervised and unsupervised segmentation methods.
• Preprocessing was demonstrated to be unnecessary for this specific saliency segmentation approach.

Conclusions:

• Image preprocessing can be omitted in saliency segmentation of skin lesions, particularly in heterogeneous dermoscopic images.
• The developed method offers a simplified yet effective approach for melanoma diagnosis support.
• This research contributes to the advancement of technology-assistive diagnostics in dermatology.