DermAI: A Full-Stack Deep Learning Web Application for Explainable Skin Lesion Screening Using MobileNetV2 and Grad-CAM

Abstract: Skin cancer, including melanoma and basal cell carcinoma, is among the most common cancers worldwide, and early visual screening is a key factor in prognosis, yet dermatologist access is limited in many regions and unaided visual assessment of pigmented lesions is known to be unreliable without specialist training. This paper presents DermAI, a full-stack web application for automated skin lesion screening built around a FastAPI backend, a Python/SQLite-free, stateless PyTorch inference pipeline, and a vanilla HTML/CSS/JavaScript front end. The system exposes REST endpoints for health checking, disease taxonomy lookup, sample-image retrieval, and image-based prediction, the last of which runs a MobileNetV2 convolutional neural network with a custom classification head designed for the seven-class HAM10000 dermatoscopic taxonomy and returns a Gradient-weighted Class Activation Mapping (Grad-CAM) visual explanation alongside every prediction. Beyond the classification and explanation pipeline, the system is architected with a dedicated clinical taxonomy module — mapping each of the seven diagnostic classes to a plain-language description, risk level, typical dermoscopic features and recommended next steps — that enriches every prediction response rather than returning a bare label. We describe the system architecture, the model and training pipeline, the disease taxonomy and REST API, the front-end design, implementation details, the testing procedure followed, safety and data considerations, a comparison against manual self-examination and generic alternatives, and the current implementation status of every module, before outlining the roadmap for completing model training, quantitative evaluation, and fairness analysis.

Keywords: Skin Lesion Classification, Convolutional Neural Network, MobileNetV2, Transfer Learning, Explainable AI, Grad-CAM, HAM10000, Computer-Aided Diagnosis, FastAPI, Teledermatology, Full-Stack Web Development.