Skin cancer is common in society, and early detection markedly enhances treatment efficacy. Deep learning-based computer-aided diagnosis systems exhibit significant potential in skin lesion classification; nevertheless, their practical use is constrained by insufficient interpretability. This research seeks to create a precise and interpretable deep learning model with the HAM10000 dataset. Three architectures, MobileNet, DenseNet, and XceptionNet, were trained and assessed, employing data augmentation to mitigate class imbalance. MobileNet attained the greatest classification accuracy at 84.04%, succeeded by DenseNet at 83.57% and XceptionNet at 81.52%. The model’s performance was evaluated using accuracy, precision, recall, F1-score, and AUC metrics. To address the critical need for transparency in deep learning-based skin cancer detection, this study integrates explainable artificial intelligence (XAI) techniques, namely Grad-CAM and LIME, to elucidate model decision-making processes. These methods highlight key regions influencing classification outcomes, significantly enhancing model interpretability. Our findings demonstrate that XAI integration not only improves the reliability and clinical utility of deep learning models but also bridges the trust gap between AI systems and healthcare practitioners, fostering their adoption in clinical practice.
ARAS et al. (Sun,) studied this question.