Development of a deep learning model for skin cancer diagnosis and its interpretation using explainable artificial intelligence techniques

Skin cancer is one of the most prevalent types of cancer worldwide, leading to severe health issues and high mortality rates if not diagnosed early. Traditional diagnostic methods are time-consuming processes requiring expertise and may exhibit reduced accuracy in some cases. In this study, deep learning models were trained for skin cancer diagnosis, and Explainable Artificial Intelligence (XAI) methods were employed to enhance the interpretability of the models' decision mechanisms. The HAM10000 dataset was utilized to classify skin lesions within the scope of the research. During the model training process, deep learning architectures such as MobileNet, DenseNet, and XceptionNet were used. The accuracy rates were obtained as 84.04% for MobileNet, 83.57% for DenseNet, and 81.52% for XceptionNet.To improve the interpretability of model predictions, Grad-CAM and LIME methods were employed. Grad-CAM highlighted the regions on which the model focused during classification through visual explanations, while LIME provided localized, instance-based explanations. This study demonstrates that AI-powered systems significantly enhance the accuracy and reliability of skin cancer diagnosis, providing substantial support to healthcare professionals during diagnostic processes. Artificial intelligence technologies have proven their effectiveness in the field of medical imaging, showcasing great potential for increasing early diagnosis rates and improving patients' quality of life. These findings represent a significant step toward promoting the widespread adoption of AI applications in clinical decision support systems, thereby enhancing the efficiency of healthcare services.