A Novel Approach to Interpret Multiclass Skin Disease Using Explainable AI

Skin diseases represent a major public health issue, where accurate diagnosis before treatment is crucial. Despite this, the complexity of multiclass skin disease classification compounded by the absence of large and comprehensive datasets, presents challenges in developing reliable diagnostic models. Current solutions are normally not explainable, resulting in healthcare professionals not trusting models. To address these challenges, a novel approach incorporating Explainable AI is proposed to improve both the interpretability and performance of skin disease classification. Using over 10,015 dermatoscopic images across 7 disease classes, Convolutional Neural Networks (CNNs), including XceptionNet and ResNet50, are applied after thorough research of existing works. The performance is then assessed through accuracy, precision, recall, and F1 score. The experimental results show that XceptionNet achieves superior accuracy with 82.2% compared to ResNet50 provided a better accuracy score of 78.2%. To enhance interpretability, Local Interpretable Model-agnostic Explanations (LIME) are incorporated as visual aids, which highlight the regions of the image that contribute most to the model’s predictions, making the predictions more comprehensible than using Black-box deep learning. This approach not only improves classification accuracy but also enhances trustworthiness through increased transparency. Future work will focus on image overlay in real-time video feed, improving data collections to make the models stronger, and designing a multiple-object occupancy deep learning model for applications with higher complexity.