max-loki/DermaliteAI-prototype

DermaLite AI is a hybrid edge-AI skin disease detection system designed for real-time, image-based analysis. [cite_start]Unlike traditional cloud-dependent tools, our solution performs fast, privacy-preserving inference directly on-device. It is specifically built to support early detection in low-resource and rural settings where constant internet # DermaLite AI: Edge-AI Skin Disease Screening prototype ## Team Details * **Team Name**: Ctrl+Alt+Del * **Team Leader**: Lokesh Kiruala * **Project**: AMD Slingshot 2026 Submission ## Project Overview DermaLite AI is a hybrid edge-AI skin disease detection system designed for real-time, image-based analysis. [cite_start]Unlike traditional cloud-dependent tools, our solution performs fast, privacy-preserving inference directly on-device. It is specifically built to support early detection in low-resource and rural settings where constant internet connectivity is unavailable. ## The Problem Most existing dermatology AI tools rely heavily on cloud servers. This creates two major issues: 1. **Privacy Risks**: Sensitive medical images must be uploaded to external servers. 2. **Accessibility**: Rural regions with low bandwidth cannot access these tools effectively. ## The Solution DermaLite AI solves these challenges by utilizing **AMD Ryzen AI** hardware for local inference. By running models locally, we ensure: * **Offline Capability**: Works in remote areas without internet. * **Privacy-First**: No mandatory image uploads to the cloud. * **Speed**: Real-time results using hardware acceleration. ## Key Features * **Real-time Classification**: Predicts disease categories from skin images. * **Risk Assessment**: Categorizes results into Low, Moderate, or High risk levels. * **Explainable AI (XAI)**: Generates Grad-CAM heatmaps to show the "attention" areas of the model. * **AMD Optimization**: Uses ONNX Runtime optimized for AMD Ryzen AI NPUs. ## Technical Stack | Component | Technology | | :--- | :--- | | **Model Architecture** | MobileNetV2 / EfficientNet-Lite | | **Framework** | PyTorch [cite: 188] | | **Optimization** | ONNX Runtime (AMD Execution Provider) | | **Interface** | Streamlit | | **Explainability** | Grad-CAM | | **Local Storage** | SQLite | ## System Architecture 1. **Frontend**: Streamlit-based local application. 2. **Preprocessing**: Image r …