DermInsight: A diagnostic System for Human Skin Diseases utilizing Deep Learning

Skin diseases present significant challenges in healthcare due to their diverse diagnosis and treatment. They ranked among the top causes of disability because most people ignore their early symptoms due to costly and time-consuming diagnostic methods, which worsen the skin condition with time. This research-based project is a contribution to the timely and cost-effective identification of seven prevailing skin diseases. We propose a novel deep learning-based system capable of classifying distinct skin diseases including melanoma, melanocytic nevi, actinic keratosis, benign keratosis, basal cell carcinoma, dermatofibroma, and vascular lesions. By leveraging a comprehensive dataset HAM10000, our system achieved an impressive accuracy rate of 98.14% in accurately identifying and categorizing these skin diseases. We employ transfer learning and fine-tuned three advance deep learning models MobileNetV1, MobileNetV2, and Xception, and evaluate their performance in the classification of seven human skin diseases. Remarkably, MobileNetV1 emerged as the top-performing model, surpassing the capabilities of the other models and existing state-of-the-art methods. In addition, the proposed model is deployed in an android app named “DermInsight” for the use of dermatologists. The dermatologist uploads the image of the skin lesions of a patient and the app will predict the disease within 1 to 2 seconds.