File size: 5,261 Bytes
e01329b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 |
# GAMMA — Glaucoma grading from Multi-Modality imAges (Challenge dataset)
<table align="center">
<tr>
<td width="100%" align="center">
<img src="rm_images/Merged_Fundus_Images_with_Captions.jpg" alt="Merged Dataset Samples" style="max-width: 100%; height: auto;">
<br>
<p><strong>Image:</strong> Dataset Samples.</p>
</td>
</tr>
</table>
## Short description
GAMMA is the first **public multi-modality glaucoma grading** dataset that pairs **2D color fundus photographs** with **3D OCT volumes** for each sample. It was released as part of the GAMMA challenge (OMIA8 / MICCAI 2021) to encourage algorithms that combine fundus and OCT information for automatic glaucoma grading.
---
## What the dataset contains
- **Paired modalities:** one macula/optic-disc centered 2D color fundus image **and** one 3D OCT volume (macula-centered) per sample.
- **Samples:** **300 paired samples** (fundus + OCT) corresponding to **276 patients**.
- **Labeling / ground truth:** each sample has a glaucoma grade (normal / early / progressive), derived from visual field mean deviation (MD) criteria; auxiliary labels include **optic disc & cup (OD/OC) segmentation masks** and **fovea coordinates** on the fundus images.
- **Demographics:** 276 Chinese patients, age range 19–77, mean ≈ 40.6 years; female ≈ 42%.
- **Balanced classes:** glaucoma ~50% of samples; within glaucoma: ~52% early, ~29% intermediate, ~19% advanced (intermediate+advanced grouped as “progressive” in challenge tasks).
- **Acquisition devices:** OCT volumes acquired using **Topcon DRI OCT Triton**; fundus images captured by **KOWA** and **Topcon TRC-NW400** cameras (macula or midpoint between disc and macula).
- **OCT spec:** 3×3 mm en-face FOV; each volume contains 256 B-scans (cross-sectional frames).
- **Image quality:** manually checked; dataset split into three challenge sets (training, preliminary, final) with ~100 pairs per set.
- **License / access:** publicly available via the GAMMA grand-challenge page; dataset distributed under **CC BY-NC-ND** (Attribution-NonCommercial-NoDerivs).
- **Official dataset page / access:** https://gamma.grand-challenge.org/
---
## Intended tasks
Primary:
- **Glaucoma grading** from paired fundus + OCT (predict: normal / early-glaucoma / progressive-glaucoma).
Auxiliary:
- **OD/OC segmentation** (optic disc and optic cup masks on fundus images).
- **Fovea localization** (x,y coordinates).
Researchers may optionally use the auxiliary tasks to boost the main grading performance.
---
## Dataset structure (typical)
```text
GAMMA/
├── images/
│ ├── fundus/ # fundus images (JPEG/PNG)
│ │ ├── sample_0001_fundus.jpg
│ │ └── ...
│ └── oct/ # OCT volumes (folder or volume files per sample)
│ ├── sample_0001_oct/ # 256 B-scans or a volume file (format described in README_original)
│ └── ...
├── labels/
│ ├── grades.csv # sample_id, grade (normal/early/progressive), MD values, other clinical metadata
│ ├── fovea_coords.csv # sample_id, x, y
│ └── od_oc_masks/ # per-sample masks (optional; may be in separate archive)
│ ├── sample_0001_od.png
│ └── ...
└── README_original.txt
```
---
## How samples were graded
Glaucoma grading ground truth was determined using **visual field mean deviation (MD)** thresholds from visual field tests performed the same day as OCT:
- **Early:** MD > −6 dB
- **Intermediate:** −12 dB < MD ≤ −6 dB
- **Advanced:** MD ≤ −12 dB
For the main challenge, intermediate + advanced were grouped as **progressive-glaucoma**.
---
## Size & splits
- **Total paired samples:** **300** (fundus + OCT)
- **Patients:** 276 (some bilateral samples)
- **Class distribution:** ~50% glaucoma / 50% non-glaucoma; within glaucoma: early ≈ 52%, intermediate ≈ 28.7%, advanced ≈ 19.3%
- **Challenge splits:** approximately **100 pairs** for training, 100 for preliminary, 100 for final test (samples from each category distributed across splits).
---
## Recommended uses & notes
- Use paired modalities (fundus + OCT) for multimodal fusion models — combining morphological cues (fundus OD/OC, vCDR) and structural OCT features (RNFL thickness) improves grading.
- Auxiliary tasks (OD/OC masks, fovea) are provided to support explainability and localized feature extraction.
- Respect the **CC BY-NC-ND** license for redistribution and commercial restrictions.
---
## Citation / sources
Please cite the GAMMA challenge paper and dataset when using the data:
- Wu J., Fang H., Li F., Fu H., Lin F., et al., **“GAMMA challenge: Glaucoma grAding from Multi-Modality imAges.”** (paper / challenge summary). arXiv:2202.06511; journal: *Medical Image Analysis* (2023). DOI: 10.1016/j.media.2023.102938.
- Official dataset page (host & download): **https://gamma.grand-challenge.org/**
Primary references used to prepare this README:
- arXiv / GAMMA challenge paper: https://arxiv.org/abs/2202.06511
- Final journal version / PubMed entry: https://pubmed.ncbi.nlm.nih.gov/37806020/
- GAMMA challenge (Grand Challenge) dataset page: https://gamma.grand-challenge.org/
--- |