Hybrid CNN–SVM Framework for Skin Cancer Detection and Stage Classification Using Simulated Hyperspectral Imaging

Abstract

Skin cancer remains one of the fastest-growing malignancies worldwide, where early detection plays a critical role in improving patient outcomes. Conventional diagnostic methods such as dermoscopy and histopathology often face challenges of subjectivity, inter-observer variability, and limited reproducibility. To address these limitations, this study proposes a hybrid Convolutional Neural Network (CNN)–Support Vector Machine (SVM) framework for skin cancer detection and stage classification using simulated hyperspectral imaging derived from dermoscopic RGB images. The methodology integrates hyperspectral simulation, preprocessing, lesion segmentation, handcrafted feature extraction based on the ABCDE rule with entropy, and CNN-based deep feature learning, followed by feature fusion and classification using SVM. Experimental evaluation on benchmark skin cancer datasets demonstrates that the hybrid CNN–SVM model achieves superior accuracy (92%), sensitivity (90%), and ROC-AUC (0.95) compared to conventional machine learning and deep learning baselines. These findings highlight the potential of the proposed framework for reliable, interpretable, and clinically relevant skin cancer diagnosis and staging.