Skin Disease Classification using Hybrid AI based Localization Approach

Although computer-aided diagnosis (CAD) is employed in a number of medical specialties, including colonography and mammography, wherein noninvasive screening procedures are carried out solely with the naked eye and there is a possibility of preventable inaccuracy, it is not utilised in dermatological observations. This paper presents a novel way to successively combine precise segmentation and classification models, demonstrating that CAD may also be a feasible choice in dermatology. We dissect a photograph of the skin in order to locate high-level characteristics and normalise the image. We first construct a segmented map of the image using a neural network-based segmentation model, after which we group the areas of aberrant skin and feed this data to a classification model. Using a separate neural network model, we categorise each cluster into various prevalent skin conditions. In comparison to other experiments, our segmentation model performs better and also gets a nearly perfect sensitivity score under challenging circumstances. Our classification model can classify numerous diseases in a single image and is more accurate than a baseline model trained without segmentation. The discipline of dermatology may be able to use CAD due to this increasing efficiency.