README.md

---
title: PRETI Retinal Disease Detection
emoji: 🔬
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 4.0.0
app_file: app.py
pinned: false
license: mit
---

# PRETI Retinal Disease Detection System

**Undergraduate Thesis — Department of Biomedical Engineering — 2025**

## What this does

This system detects 4 retinal diseases simultaneously from a single fundus photograph:
- **DR** — Diabetic Retinopathy
- **Glaucoma**
- **HR** — Hypertensive Retinopathy
- **RVO** — Retinal Vein Occlusion

It also demonstrates **AGPT (Attention-Guided Patch Transmission)** — a novel mechanism that uses PRETI's RAAM attention maps to select only disease-relevant image patches for transmission, achieving **70.3% bandwidth reduction** for rural telemedicine.

## Results

| Disease | AUC |
|---|---|
| DR | 0.9869 |
| Glaucoma | 0.9999 |
| HR | 0.9881 |
| RVO | 0.9864 |
| **Macro** | **0.9903** |

## References

- PRETI: Lee et al., arXiv:2505.12233, 2025
- Focal Loss: Lin et al., IEEE TPAMI, 2020
- Class-Balanced Sampler: Cui et al., CVPR 2019
- ViT: Dosovitskiy et al., ICLR 2021