Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
(ReadTimeoutError("HTTPSConnectionPool(host='huggingface.co', port=443): Read timed out. (read timeout=10)"), '(Request ID: 426d3fb5-cd6a-4e83-86d0-d27b8700d9c8)')
Error code:   UnexpectedError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

image
image
End of preview.

GAMMA β€” Glaucoma grading from Multi-Modality imAges (Challenge dataset)

Merged Dataset Samples

Image: Dataset Samples.

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 β‰ˆ 40.6 years; female β‰ˆ 42%.
  • Balanced classes: glaucoma ~50% of samples; within glaucoma: ~52% early, ~29% intermediate, ~19% advanced (intermediate+advanced grouped as β€œprogressive” 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: 3Γ—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)

GAMMA/
β”œβ”€β”€ images/
β”‚ β”œβ”€β”€ fundus/ # fundus images (JPEG/PNG)
β”‚ β”‚ β”œβ”€β”€ sample_0001_fundus.jpg
β”‚ β”‚ └── ...
β”‚ └── oct/ # OCT volumes (folder or volume files per sample)
β”‚ β”œβ”€β”€ sample_0001_oct/ # 256 B-scans or a volume file (format described in README_original)
β”‚ └── ...
β”œβ”€β”€ labels/
β”‚ β”œβ”€β”€ grades.csv # sample_id, grade (normal/early/progressive), MD values, other clinical metadata
β”‚ β”œβ”€β”€ fovea_coords.csv # sample_id, x, y
β”‚ └── od_oc_masks/ # per-sample masks (optional; may be in separate archive)
β”‚ β”œβ”€β”€ sample_0001_od.png
β”‚ └── ...
└── 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 > βˆ’6 dB
  • Intermediate: βˆ’12 dB < MD ≀ βˆ’6 dB
  • Advanced: MD ≀ βˆ’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 β‰ˆ 52%, intermediate β‰ˆ 28.7%, advanced β‰ˆ 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 β€” 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., β€œGAMMA 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:


Downloads last month
5