SKIN MAP: A multimodal approach to detect and explain skin cancer

Early and reliable detection of skin cancer remains a clinical challenge due to subtle visual differences between benign and malignant lesions and the limited interpretability of deep learning models. This work presents SkinMap, a multimodal system that combines dermoscopic images with patient metadata for improved skin-lesion classification. Our approach uses a deep convolutional model to extract hierarchical image features and integrates structured clinical attributes for binary and multiclass prediction tasks. To enhance transparency, we incorporate Grad-CAM-based visual explanations that highlight diagnostically relevant regions and support clinical reasoning. This framework aims to improve diagnostic accuracy while increasing trust in AI-assisted dermatological assessment.