EARLY-STAGE DISEASE DETECTION USING FACIAL IMAGE ANALYSIS WITH DEEP LEARNING AND GENAI
DOI:
https://doi.org/10.62643/ijerst.2026.v22.n2.pp304-308Keywords:
Skin Disease Classification; EfficientNet-B3; Transfer Learning; Vision-Language Model; DermNet; FastAPI; Streamlit; Medical AI; Top-K Accuracy; DermatologyAbstract
Automated dermatological diagnosis using deep learning has emerged as a promising complement to clinical assessment. This paper presents a two-stage intelligent pipeline for skin disease detection: an EfficientNet-B3 convolutional classifier trained on 19,582 images from the DermNet NZ dataset across 21 disease categories, followed by a Google Gemini 2.5 Flash vision-language model (VLM) that generates interpretable, patient-friendly medical explanations for each prediction. Trained for 40 epochs with AdamW optimisation and CosineAnnealingLR scheduling, the classifier achieves a Top-1 test accuracy of 68.31%, Top-3 accuracy of 84.98%, and Top-5 accuracy of 89.99%, with a macro-averaged F1-score of 0.638. A production-ready system comprising a FastAPI backend and Streamlit frontend delivers real-time predictions with confidence scores and VLM-generated explanations, demonstrating the practical feasibility of AI-assisted dermatology screening.
Downloads
Published
Issue
Section
License

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.













