Spaces:
Sleeping
Sleeping
| title: LCVC-Ensemble Brain Tumor MRI Classifier | |
| sdk: gradio | |
| app_file: app.py | |
| pinned: false | |
| license: mit | |
| # LCVC-Ensemble Brain Tumor MRI Classifier | |
| This Space contains a research demonstration of **LCVC-Ensemble** | |
| (**Leakage-Controlled Validation-Calibrated Multi-Backbone Ensemble**) for four-class brain tumor MRI classification. | |
| ## Classes | |
| glioma, meningioma, notumor, pituitary | |
| ## Test metrics from the leakage-aware evaluation protocol | |
| - Accuracy: 0.9887640449438202 | |
| - Macro-F1: 0.9887621278421996 | |
| - Balanced accuracy: 0.9887638076313365 | |
| - Macro-AUC OVR: 0.9997131229291261 | |
| ## Ensemble | |
| The app loads `5` selected checkpoints and performs weighted averaging of temperature-scaled softmax probabilities. | |
| ## Intended use | |
| Research and educational demonstration only. | |
| ## Medical disclaimer | |
| This is **not a medical device**. It must not be used for diagnosis, treatment, triage, or clinical decision-making. | |
| ## Reproducibility note | |
| The ensemble was selected using validation Macro-F1 only. Test labels were used only for final reporting after model selection. | |