usama10/retinal-dr-longitudinal

---
license: cc-by-nc-4.0
task_categories:
- image-to-image
language:
- en
tags:
- diabetic-retinopathy
- retinal-fundus
- longitudinal
- medical-imaging
- ophthalmology
- paired-images
- disease-progression
size_categories:
- 1K<n<10K
---

# Retinal DR Longitudinal: Paired Baseline and 2-Year Follow-up Fundus Images

Paired retinal fundus photographs from 574 diabetic retinopathy (DR) patients, with baseline and two-year follow-up images plus clinical metadata.

## Dataset Description

This dataset contains longitudinal retinal fundus image pairs acquired at Tianjin Medical University Eye Hospital. Each patient has a baseline fundus photograph and a corresponding two-year follow-up photograph, enabling research on disease progression prediction, longitudinal image synthesis, and temporal retinal change analysis.

**Ethics:** This study adhered to the Declaration of Helsinki and was approved by the institutional review board of Tianjin Medical University Eye Hospital (approval number 2019KY-22). Written informed consent was obtained from all participants.

## Dataset Statistics

| Property | Value |
|----------|-------|
| Patients | 574 |
| Baseline image folders | 624 |
| Follow-up images | 1,178 |
| Usable paired observations | 1,148 |
| Baseline image resolution | 2992 x 2000 px |
| Follow-up image resolution | 3472 x 2320 px |
| DR progression rate | 19.2% |
| Image format | JPEG |

## Clinical Metadata

The file `Organized_Data of Patients.xlsx` contains two sheets (Baseline, 2-Year Follow-up) with the following variables per patient:

| Variable | Type | Description |
|----------|------|-------------|
| Patient ID | Identifier | Unique patient number |
| Age | Continuous | Patient age in years (mean 63.2, SD 8.1) |
| Sex | Binary | Male / Female (47.2% male) |
| HbA1c (%) | Continuous | Glycated hemoglobin (mean 7.3, SD 1.2) |
| Fasting glucose | Continuous | Fasting blood glucose (mg/dL) |
| DR grade | Ordinal | ETDRS severity scale (1-5) |
| Hypertension | Binary | Hypertension status (42.8% hypertensive) |

Grades 6 (post-photocoagulation) and 7 (ungradable) are present in the metadata but should be excluded from modeling due to non-standard retinal appearance or data incompleteness.

## Directory Structure

```
baseline fundus images/
00194/
00194-7256.jpg # Left or right eye
00194-7261.jpg # Other eye (if available)
00197/
...
(624 patient folders)

2 year follow-up fundus images/
1036_2.jpg # Flat directory, patient ID in filename
1036_4.jpg
...
(1,178 images)

Organized_Data of Patients.xlsx # Clinical metadata
```

## Matching Baseline to Follow-up

Baseline images are organized in folders named by patient ID. Follow-up images are in a flat directory with filenames containing the patient ID as the prefix before the underscore. Some follow-up filenames have timestamp suffixes (`_YYYYMMDD_HHMMSS`) that should be stripped when parsing. Some baseline filenames have leading-zero inconsistencies; match by folder name rather than filename.

## Recommended Data Split

We recommend an 80/10/10 train/validation/test split at the **patient level** (both eyes of the same patient must be in the same split), stratified by DR progression label.

## Associated Paper

Muhammad Usama, Emmanuel Eric Pazo, Xiaorong Li, Juping Liu. "Conditional Latent Diffusion for Predictive Retinal Fundus Image Synthesis from Baseline Imaging and Clinical Metadata." *Computers in Biology and Medicine* (under review), 2026.

Code: [github.com/Usama1002/retinal-diffusion](https://github.com/Usama1002/retinal-diffusion)

Model weights: [huggingface.co/usama10/retinal-diffusion-model](https://huggingface.co/usama10/retinal-diffusion-model)

## Citation

```bibtex
@article{usama2026retinal,
title={Conditional Latent Diffusion for Predictive Retinal Fundus Image Synthesis from Baseline Imaging and Clinical Metadata},
author={Usama, Muhammad and Pazo, Emmanuel Eric and Li, Xiaorong and Liu, Juping},
journal={Computers in Biology and Medicine (under review)},
year={2026}
}
```

## License

This dataset is released under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). It may be used for non-commercial research purposes only. Redistribution of the raw images outside this repository is not permitted without explicit permission from the data providers.