Diagnostic techniques for improved segmentation, feature extraction, and classification of malignant melanoma

Area of Science:

• Dermatology
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Malignant melanoma diagnosis relies on image segmentation, feature extraction, and classification.
• Existing methods face limitations in accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate novel methods for melanoma image segmentation, feature extraction, and classification.
• To improve the diagnostic accuracy of distinguishing malignant from benign melanomas.

Main Methods:

• Replaced edge-imfill with U-Otsu method for image segmentation.
• Introduced new features for the ABCD criteria (asymmetry, border irregularity, color variegation, diameter).
• Implemented weighted receiver operating characteristic (ROC) thresholding for classification, replacing median thresholding.

Main Results:

• The suggested segmentation method showed remarkable improvement compared to the previous method, validated against expert segmentation.
• New feature extraction and classification methods also demonstrated significant enhancements in performance metrics.
• Overall, all three proposed steps resulted in substantial improvements in diagnostic accuracy, sensitivity, and specificity.

Conclusions:

• The novel U-Otsu segmentation, ABCD feature extraction, and weighted ROC classification methods offer a significant advancement in melanoma diagnosis.
• These improved techniques enhance the ability to accurately differentiate between malignant and benign skin lesions.
• The proposed approach shows great potential for clinical application in early and accurate melanoma detection.