DermAI : AI Based Skin Disease Classifier

Abstract

 Skin diseases represent one of the global health issues with high prevalence rates, and early identification by appropriate diagnosis can help prevent complications and increase success rates. However, conventional skin disease diagnosis by dermatologists is a highly intensive and subjective process, especially in remote regions with scarce dermatological knowledge and skills. As such, this study proposes an automated skin disease classification systemäs based on EfficientNet-B0). This proposed model uses dermoscopic and clinical skin images and requires resizing, normalization, and data augmentation as part of its image preprocessing techniques to enhance its generalizability and robustness levels. Additionally, despite using skin lesion images for training and identifying their corresponding essential features using transfer learning techniques for training, this proposed automated skin disease classification system has demonstrated capabilities to significantly improve accuracy rates at lower precision, recall, and F1-score values compared to conventional CNN models and also provide faster training and inference speeds in its experimentation and validation phases. This also demonstrates that it would be highly beneficial for use in remote regions where skin lesion image classification can be performed in real-time due to its lightweight capabilities and high accuracy rates offered by EfficientNet-B0).