Interpretable AI for Skin Screening: MultiExCam for Melanoma, Basal Cell Carcinoma and Benign Nevi Classification

Accurate differentiation among melanoma, basal cell carcinoma (BCC), and benign nevi remains challenging in clinical dermatology, despite their distinct prognostic implications. While recent advances in artificial intelligence have improved binary skin lesion classification, multiclass scenarios reflecting real-world diagnostic complexity remain still challenging. This study extends MultiExCam, a hybrid deep learning and machine learning framework, to address multiclass classification with enhanced explainability for clinical decision support. By incorporating basal cell carcinoma as a third diagnostic class, we demonstrate how hybrid integration of deep and machine learning can effectively address real-world multiclass skin lesion scenarios. Explainability analysis via Grad-CAM and SHAP enables identification of clinically meaningful features distinguishing melanoma (asymmetry, irregular borders, color variation), BCC (vascular patterns, pearly appearance, ulceration), and nevi (symmetry, homogeneity), providing interpretable decision rationales aligned with dermatological diagnostic criteria. The framework’s ability to differentiate among three distinct lesion types with transparent decision-making processes addresses critical requirements for clinical adoption, offering a promising foundation for AI-assisted dermatology and screening applications in resource-limited settings.