fairdataihub/envision-eye-imaging-training-data

---
license: mit
tags:
- text-classification
- eye-imaging
- ophthalmology
- medical-imaging
- fair-data
- eyeact
---

# ENVISION Eye Imaging Training Data

Training dataset for the [ENVISION eye imaging classifier](https://huggingface.co/fairdataihub/envision-eye-imaging-classifier).

## Dataset Description

994 curated text examples for binary classification of eye imaging datasets from scientific metadata.

| Label | Count | Description |
|-------|-------|-------------|
| EYE_IMAGING | 365 | Ophthalmic imaging datasets (fundus, OCT, OCTA, cornea) |
| NEGATIVE | 629 | Non-imaging data (software, other eye research, unrelated) |

### Source Composition

**EYE_IMAGING (365)**:
- 77 hand-curated examples from multi-repository sources
- 288 expert-verified positives from cross-repository classification

**NEGATIVE (629)**:
- 502 general negative examples
- 48 eye-related software examples (hard negatives)
- 79 non-imaging eye data examples (hard negatives)

### Data Format

Each example is a text string representing concatenated metadata fields (title, description, keywords, file types) from scientific data repositories (Zenodo, Figshare, Dryad, Kaggle, NEI).

## Usage

```python
from datasets import load_dataset

ds = load_dataset("fairdataihub/envision-eye-imaging-training-data")
```

## Citation

- EyeACT Envision project
- FAIR Data Innovations Hub (fairdataihub.org)