---
title: DeepSeeNet
emoji: 🐢
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 6.14.0
python_version: '3.13'
app_file: app.py
pinned: false
license: mit
short_description: Framework for Classifying patient-based AMD in CFP images
---

# DeepSeeNet PyTorch

This repository is a PyTorch reimplementation of the original DeepSeeNet model:

https://github.com/ncbi-nlp/DeepSeeNet

DeepSeeNet predicts patient-level AREDS Simplified Severity Scale scores for age-related macular degeneration (AMD) from bilateral color fundus photographs. The model follows the original DeepSeeNet design by first predicting eye-level AMD risk factors, then combining predictions from both eyes into a patient-level simplified severity score.

## Tasks

The implementation trains three image-level subnetworks:

| Task | Classes | Output |
|---|---:|---|
| `ADVAMD` | 2 | late AMD absent / present |
| `DRUS` | 3 | small/none, medium, large drusen |
| `PIG` | 2 | pigmentary abnormality absent / present |

The final AREDS simplified score is computed from bilateral predictions:

- score `5` if late AMD is predicted in either eye
- otherwise, score is based on large drusen and pigmentary abnormalities across both eyes
- bilateral medium drusen contributes one point

## Citation

If you use this repository, please cite the original DeepSeeNet paper:

```bibtex
@article{peng2019deepseenet,
  title={DeepSeeNet: A Deep Learning Model for Automated Classification of Patient-based Age-related Macular Degeneration Severity from Color Fundus Photographs},
  author={Peng, Yifan and Dharssi, Shazia and Chen, Qingyu and Keenan, Tiarnan D. and Agr\'{o}n, Elvira and Wong, Wai T. and Chew, Emily Y. and Lu, Zhiyong},
  journal={Ophthalmology},
  volume={126},
  number={4},
  pages={565--575},
  year={2019},
  publisher={Elsevier},
  doi={10.1016/j.ophtha.2018.11.015}
}
```