Enhanced Skin Cancer Diagnosis Using a CAD System with CNN Feature Extraction and GKSO-Optimized Classification

Early detection of skin cancer significantly improves treatment success, reduces mortality, and enhances patient outcomes. Traditional detection procedures, such as visual inspection and biopsy, are time-consuming, subjective, and reliant on specialist dermatologists, resulting in diagnostic delays. A Computer-Aided Diagnosis (CAD) system enhances early detection with the help of advanced image processing and machine learning (ML) techniques to improve accuracy and reduce diagnosis time. CADbased systems typically involve data acquisition, preprocessing, feature extraction, selection, and classification. However, the high dimensionality of skin cancer datasets can degrade the ML model performance by introducing noise and irrelevant features. Feature selection is crucial for identifying key features, reducing complexity, and improving classification accuracy. This study proposes a new CAD system for skin cancer detection based on pretrained CNN models for feature extraction, Genghis Khan Shark Optimizer (GKSO) for feature selection, and K-Nearest Neighbors for classification. Experimental results reveal that the proposed CAD system significantly improves diagnostic accuracy.