Skin Disease Classification using Hybrid Framework and Automated Clinical Insights using LLM

Millions of people worldwide suffer from skin conditions that have an impact on their physical, mental, and financial well-being. These conditions include bacterial, fungal, viral, and parasitic infections. Timely actions depend on accurate classification and early diagnosis. The paper discusses the classification of eight common skin diseases through a hybrid deep learning method that integrates ResNet, MobileNetV3Small, and VGG16. The integration offers high accuracy (94.12%) in dermatoscopic image classification, useful in cases of limited dermatological access. Besides diagnosis, the paper also features another information module based on Large Language Models (LLMs) like Gemini, ChatGPT, and Meta Copilot. The models generate medical descriptions, cross-verified against authoritative sources using cosine similarity metrics. Gemini was the most accurate model, with 71% accuracy and a perplexity score of 2.46. These two systems enhance dermatological AI by integrating deep learning for diagnosis and NLP-based medical guidance, which ensures a holistic approach to skin disease management.