MSCC: Multi-Class Skin Cancer Classification and Interpretable Deep Learning Systems

Nowadays, skin cancer is a common and potentially deadly disease, requiring prompt and precise diagnosis for effective treatment. Our study introduces a multi-class skin cancer classification (MSCC) system employing deep convolutional neural networks (DCNNs) and interpretable deep learning frameworks. This approach enhances accuracy and speed, providing clear, understandable visual explanations. By addressing challenges like feature extraction from irregular, artifactladen images and improving generalization and interpretability, this system promises to significantly aid in early detection, thus saving lives and reducing the strain on healthcare professionals. The effectiveness of the proposed model is assessed using the ISIC-2018 and ISIC-2019 datasets for skin imaging. The model successfully distinguishes between seven types of skin lesions-benign ker-atosis lesions, melanoma, basal cell carcinoma, melanocytic nevi, vascular lesions, actinic keratosis, and dermatofibroma-with high accuracy, precision, recall, and F1 score, all averaging at 96.37%, 96.39%, 96.35%, and 96.36%, respectively. To delve deeper into the model's predictions, we employ the local in-terpretable model-agnostic explanations (LIME) framework and SHapley Additive exPlanations (SHAP) values. These techniques generate visual explanations aligned with prior beliefs and adhere to best practices for general explanations. The incorporation of explainability enhances the model's utility in real clinical scenarios.