Standardized Multi-Channel Dataset for Glaucoma (SMDG-19), a standardization of 19 public datasets for AI applications.
This dataset is now available on Kaggle! Check it out!
Standardized Multi-Channel Dataset for Glaucoma (SMDG-19) is a collection and standardization of 19 public full-fundus glaucoma images, associated image metadata like, optic disc segmentation, optic cup segmentation, blood vessel segmentation, and any provided per-instance text metadata like sex and age. Standardized images like those presented here are not only machine-learning ready, but much more portable, useable, and accessible than their original formats. For example, the popular dataset, EyePACS AIROGS, contains nearly 100GB of data without standardization and around 5GB of data after standardization!
Accessibility
Dataset Link: https://www.kaggle.com/datasets/deathtrooper/multichannel-glaucoma-benchmark-dataset
Dataset Objective
The objective of this dataset that is a machine learning-ready dataset for Glaucoma-related applications. Using the help of the community, new open-source Glaucoma datasets will be reviewed for standardization and inclusion in this dataset.
Public Datasets
Learn more about existing public glaucoma datasets using this dataset catalog: https://github.com/TheBeastCoding/glaucoma-dataset-metadata
Dataset Objective
The objective of this dataset that is a machine learning-ready dataset for Glaucoma-related applications. Using the help of the community, new open-source Glaucoma datasets will be reviewed for standardization and inclusion in this dataset.
Data Standardization
- Full fundus images (and corresponding segmentation maps) are standardized by cropping the background, centering the fundus image, padding missing information, and resizing to 512x512 pixels.
- Each available metadata text attribute is provided as a column in a CSV file
| Dataset Instance | Original Fundus | Standardized Fundus Image |
|---|---|---|
| sjchoi86-HRF |  |
 |
| BEH |  |
 |
Dataset labels
- 0: Non-Glaucoma instance
- 1: Glaucoma instance
- -1: Glaucoma-suspect instance
SMDG-19 is comrpised of the following Public Glaucoma Image Datasets
Instructions for Popular Use Cases
- Glaucoma classification (12,449 total instances): Split the data by 'types' column in the CSV file. Input = 'fundus' file. Label = 'types' number.
- Optic cup segmentation (2,874 instances): Find all rows in CSV file with a non-empty 'fundus_od_seg' column. Input = 'fundus' file. Label = 'fundus_oc_seg' file.
- Optic disc segmentation (3,103 instances): Find all rows in CSV file with a non-empty 'fundus_oc_seg' column. Note some instances are labeled as 'Not Visible', so you must exclude these as well. Input = 'fundus' file. Label = 'fundus_od_seg' file.
- Blood vessel segmentation (462 instances): Find all rows in CSV file with a non-empty 'bv_seg' column. Input = 'fundus' file. Label = 'bv_seg' file.
File Descriptions
- metadata.csv : Links dataset instance metadata to image file paths.
- full-fundus/ : Folder containing all full fundus images.
- optic-cup/ : Folder containing the optic cup segmentation map based on the full fundus image.
- optic-disc/ : Folder containing the optic disc segmentation map based on the full fundus image.
- blood-vessel/ : Folder containing the blood vessel segmentation map based on the full fundus image.
- vessel-artery/ : Folder containing the artery segmentation map based on the full fundus image.
- vessel-vein/ : Folder containing the vein segmentation map based on the full fundus image.
- spectral-oct/ : Folder containing all full spectral oct images.
- spectral-oct-cup/ : Folder containing the optic cup segmentation lines based on the full spectral oct image.
- spectral-oct-disc/ : Folder containing the optic disc segmentation lines based on the full spectral oct image.
The following datasets are open-access but do not contain full fundus images. These images are not included in our dataset
- ACRIMA
- KEH (Kim's Eye Hospital)
- RIM-ONE