Exploring Explainable AI in Medical Image Segmentation: A Case Study on Skin Disease Detection

It is crucial to identify and classify skin diseases so that the right treatment can be administered at the right time. The present work aims to explore the existence of deep learning, especially the U-Net model and its derivatives, in enhancing skin disease detection through image segmentation. In our study, a large dataset of dermatoscopic images is used to assess the efficiency of the U-Net, DeepLabV3+, and U-Net3+ architectures. We also consider the need to incorporate Explainable AI (XAI) approaches to promote trust and understanding of these models. The saliency map and Grad-CAM are some of the explainable artificial intelligence methods that help in enhancing clinical validation to understand the model’s decision-making process. From the results presented, it can be concluded that the improvements are more noticeable for advanced U-Net variants, especially for U-Net3+ while using the same architectural backbone as for the baseline U-Net but with added measures of robustness. These architectures, when integrated with XAI techniques, may serve a significant purpose in the development of automated systems to identify skin diseases at an early stage and with high accuracy that will help patients and medical professionals. The focus of the subsequent studies will be on integrating these models into an instantaneous diagnosis tool; and on expanding the given set of skin diseases.