A Mobile-Based Deep Learning System for Skin Disease Diagnosis

A proper and timely diagnosis of dermatological conditions is critical for maximizing therapy efficacy and improving patient outcomes. Traditional diagnostic methods frequently rely on dermatologists' subjective and time-consuming visual assessments. This paper focuses on addressing challenges in healthcare by developing a mobile-based system for diagnosing skin diseases using a mobile camera and deep learning architecture. The front-end and back-end are the two primary components of the proposed mobile-based deep learning system. In the front-end part, the user used the mobile camera to capture the damaged skin area. Then, the captured image is used to feed the back-end part. In this part, pre-processing methods are followed by applying the proposed modified ResNet50 architecture for the captured image. Further processing layers are added to improve the ResNet50's ability to distinguish between five distinct kinds of skin diseases. Data oversampling and data augmentation techniques are the main components of the pre-processing phase. Then, based on the classification result, the user receives a detailed report with the diagnosis and recommended treatment. The experimental results indicate that the proposed system effectively distinguishes lupus, urticaria, cancer, and burns from normal skin tissues with significant accuracy. The proposed system-based modified ResNet50 achieved an accuracy of 90.74%, a precision of 87.78%, a sensitivity of 86.84%, and a f1-score of 87.04% F1-score. Furthermore, the results revealed that the proposed system exhibits remarkable competitiveness when compared to state-of-the-art models.