Computer-aided diagnosis of melanoma using border and wavelet-based texture analysis.

Area of Science:

• Medical image analysis
• Computational dermatology
• Machine learning in healthcare

Background:

• Melanoma diagnosis relies heavily on visual inspection of skin lesions.
• Computer-aided diagnosis (CAD) systems offer potential for improved accuracy and efficiency.
• Integrating diverse lesion features can enhance diagnostic performance.

Purpose of the Study:

• To develop and evaluate a novel CAD system for melanoma diagnosis.
• To optimize the selection and integration of textural, border-based, and geometrical features.
• To assess the performance of different classifiers for melanoma classification.

Main Methods:

• Feature extraction using wavelet decomposition (texture), boundary series modeling (border), and shape indexes (geometry).
• Optimized feature selection via the computationally efficient Gain-Ratio method.
• Classification using Support Vector Machine, Random Forest, Logistic Model Tree, and Hidden Naive Bayes.

Main Results:

• The proposed system achieved an accuracy of 91.26% and an Area Under the Curve (AUC) of 0.937 on a dataset of 289 dermoscopy images.
• Combining texture, border, and geometry features significantly outperformed using texture features alone.
• Texture features demonstrated a higher contribution than border-based features in the optimized feature set.

Conclusions:

• The developed CAD system demonstrates high accuracy for melanoma diagnosis.
• The integration of multi-modal features (texture, border, geometry) is crucial for enhancing diagnostic performance.
• Optimized feature selection and integration are key to developing effective CAD systems for melanoma.