XAAI-ledger: An explainable CNN-transformer-based multi-modal deep learning framework for early detection of melanoma and non-melanoma skin cancers using dermoscopic and clinical data

Over the past few decades, skin cancer—particularly melanoma—has become a significant global health problem due to its rising incidence and challenging diagnosis. It is believed that improving patient outcomes requires both accuracy and early detection. This paper presents XAAI-Ledger, an explainable and hybrid Convolutional Neural Network (CNN) Transformer-enabled multi-modal deep learning framework that combines clinical metadata and dermoscopic images to categories skin cancers, including melanoma and non-melanoma. In addition to using CNN to extract spatial features, the Transformer Encoder (TE) in this proposed solution aids in collecting semantic linkages and long-range dependencies across medical record modalities. To improve interpretability and decision accuracy, visual and non-visual elements are combined using an attention-based fusion method. Explainable AI (XAI) mechanisms, such as SHapley Additive exPlanations (SHAP) and Gradient-Weighted Class (Grad-CAM) Activation Mapping, are utilized to foster therapeutic trust. These mechanisms not only make it possible to visualize model thinking but also offer efficacy and dependability. Simulation findings on multiple benchmark datasets, particularly ISIC 2019, show a 98.71% classification accuracy, a 97.45% sensitivity, and a 98.22% specificity. When compared to alternative cutting-edge techniques, it takes into account the superior metrics that this proposed model achieves. In summary, by offering a transparent, resilient, and high-performance solution that is appropriate for dynamic clinical environments, the proposed XAAI-Ledger advances AI-enabled dermatological diagnostic procedures.