An Interpretable Multimodal Approach for Skin Cancer Analysis via Simulated Hyperspectral Reconstruction and Hybrid Machine Learning

Skin cancer diagnosis requires accurate lesion classification and reliable stage estimation to enable timely clinical intervention. This study proposes a stage-aware multimodal framework integrating simulated hyperspectral reconstruction with hybrid deep learning and machine learning techniques. Initially, RGB dermoscopic images are preprocessed and enhanced, followed by feature-level class balancing using SMOTE. Principal Component Analysis (PCA) is employed to generate simulated hyperspectral representations, enriching spectral information beyond conventional RGB imaging. Deep semantic features are extracted using a Convolutional Neural Network (CNN), while clinically significant handcrafted features—including tumor diameter, asymmetry index, circularity, and entropy—are computed to capture morphological characteristics. These features are fused and classified using a Support Vector Machine (SVM) for tumor detection, followed by Artificial Neural Network (ANN)-based stage estimation. Experimental results demonstrate that the proposed hybrid framework achieves superior performance, with improved classification accuracy, enhanced robustness under class imbalance, and better generalization compared to single-model approaches. Furthermore, a stage-aware clinical recommendation module provides interpretable and actionable diagnostic insights, enhancing real-world applicability. The proposed system offers a cost-effective and explainable solution for intelligent skin cancer diagnosis.