Enhanced Skin Disease Classification and Segmentation with U-Net and Multilayer CNN

Skin diseases, encompassing various cancers and dermatological conditions, are the leading health problems worldwide. Early diagnosis is essential for effective management and prevention of complications. Still, the existing diagnostic approaches, such as manual assessment and traditional machine learning, face problems of imprecision and inefficiency because of fewer data augmentations, inadequacy in segmentation methods, and suboptimal feature extraction. This paper proposes a robust framework, combining advanced deep learning approaches to overcome these limitations. Variational Autoencoders are used for data augmentation, and the U-Net network provides accurate segmentation. A user-defined multilayer Convolutional Neural Network (CNN) is utilized to identify 23 skin conditions from more than 19,500 high-quality dermoscopic images. The proposed model utilizes sequential feature extraction, increased network depth, and efficient regularization, thereby attaining 97.30% accuracy, surpassing earlier studies reporting 82-89% accuracy. It also has real-time diagnostic explanations and offers delay and healthcare cost savings. Findings emphasize the potential of deep learning to transform dermatology by allowing for accurate, accessible, and efficient early diagnosis of skin diseases.