Harnessing deep learning and explainable AI for accurate skin rash detection and classification

Abstract

Skin diseases have become a growing issue all over the world, skin rashes are one of the most common among them. This paper focuses on establishing a viable deep learning approach to distinguish between skin rashes. For this purpose, Deep Neural Network Models, such as CNN, MobileNetV2, and DenseNet201 algorithms, were implemented and analyzed for suitability in this kind of image data set. In this research, a synthetic dataset consisted of 1381 images of three distinct skin rash diseases, which were pre-processed using the image resizing and dataset augmentation process. Among the algorithms implemented, multiple algorithms gave high accuracy scores, which suggested the high quality of the data set. The models were also interpreted using an Explainable AI technique named LIME. MobileNetV2 provided the highest precision among the algorithms implemented, followed by DenseNet201, indicating the viability of using DNN architectures in skin rash detection.