Skin Disease Detection and Severity Estimation Using Hybrid Resnet-50 and U-Net Architectures

–Skin diseases is one of the most crucial challenges and with its early detection depends on dermatologist expertise. Nonetheless, not all people have access to healthcare thereby creating delays in diagnosis. This research aims to produce an AI based model such that it can automatically detect skin diseases and measuring the severity of the disease using deep learning. The model combines ResNet-50 and U-Net classes where ResNet-50 is used to classify disease and U-Net to segment the cancerous regions. The users are allowed to upload images of the skin and the system shall predict the type of disease and the degree of it mild, moderate, severe. This research aims to train a CNN model with optimized layer configuration and data augmentation, with batch size of 60, running over 400 epochs, finetuned to fit all cases with an accuracy of 97.79% and an F1 score of 98% showing it robustness.