| title: AI-Powered Diabetic Retinopathy Grading System from Retinal Fundus Images Using Deep Learning | |
| emoji: 👁️ | |
| colorFrom: blue | |
| colorTo: indigo | |
| sdk: gradio | |
| sdk_version: 5.29.0 | |
| app_file: app.py | |
| pinned: false | |
| license: mit | |
| # Diabetic Retinopathy Grading AI (v9) | |
| An AI-powered clinical diagnostic tool for grading Diabetic Retinopathy from retinal fundus photographs. | |
| Powered by our latest **v9** training phase. | |
| ## Performance Metrics (v9) | |
| - **Validation QWK**: 0.9167 | |
| - **Test QWK**: 0.8954 | |
| - **Binary Referable-DR Accuracy**: 91.92% | |
| - **Test AUROC (macro)**: 0.8885 | |
| ## Features | |
| - **5-Grade ICDR Classification** using EfficientNetV2-M + CORAL ordinal regression | |
| - **Grad-CAM++ Visual Explanations** highlighting lesion regions | |
| - **Fundus Gating** — rejects non-retinal images automatically | |
| - **4-View TTA** (Test-Time Augmentation) for improved robustness | |
| - **Premium Material Design 3 UI** with real-time processing metrics | |