Dermatology has evolved dramatically as a result of explainable artificial intelligence (XAI), especially in the early diagnosis and identification of skin diseases. The study and understanding of dermatology has long been crucial to the medical field because skin diseases are challenging to detect because of their intricate classifications. Although the accurate identification of skin diseases is a complex process, even for the most experienced dermatologists. Deep learning’s evolution into dermatological and diagnostic processes promises potential solutions to the limitations of conventional techniques, which often depend on clinical examinations. This research developed a novel architecture based on an attention module and inception blocks that works well for detecting 57 skin diseases. We make use of the dataset for cancer and skin diseases, although it only has 888 samples. To evaluate the proposed model, the dataset is preprocessed, normalised, and enhanced using augmentation techniques to improve the skin samples and avoid overfitting. Experimental results demonstrate that the proposed model 98.63% accurately detects skin diseases with a 0.062 validation loss, and a 0.0027 standard deviation. The proposed model also achieved 98.61%, 98.91%, and 98.69% macro precision, recall, and F1-score. The proposed model is compared with the most recent state-of-the-art studies in the literature, primarily from the last 2 years, 2023 and 2024. In addition, five-fold cross-validation and statistical analysis (paired t-tests) demonstrate consistent performance and statistically significant improvements. Exploring the use of the proposed XAI models in a smart healthcare environment will consider gradient weighted class activation mapping (Grad-CAM), outperforming local interpretable model agnostic explanations (LIME) for trustworthiness and cohesion. The research we conducted ensures dependability and objectivity in AI-driven decisions in a trustworthy healthcare system. Grad-CAM highlighted the important regions of the skin to improve knowledge of the predictions. The saliency maps might show the forecasts for the artificial intelligence model as a heatmap.
Alamri et al. (Wed,) studied this question.