Skip to content

Latest commit

 

History

History
80 lines (65 loc) · 6.77 KB

README.md

File metadata and controls

80 lines (65 loc) · 6.77 KB
Raw

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

Dataset 0 1 -1 Access Link
BEH (Bangladesh Eye Hospital) 463 171 0 https://github.com/mirtanvirislam/Deep-Learning-Based-Glaucoma-Detection-with-Cropped-Optic-Cup-and-Disc-and-Blood-Vessel-Segmentation/tree/master/Dataset
CRFO-v4 31 48 0 https://data.mendeley.com/datasets/trghs22fpg/4
DR-HAGIS (Diabetic Retinopathy, Hypertension, Age-related macular degeneration and Glacuoma ImageS) 0 10 0 https://personalpages.manchester.ac.uk/staff/niall.p.mcloughlin/
DRISHTI-GS1-TRAIN 18 32 0 https://cvit.iiit.ac.in/projects/mip/drishti-gs/mip-dataset2/Home.php
DRISHTI-GS1-TEST 13 38 0 https://cvit.iiit.ac.in/projects/mip/drishti-gs/mip-dataset2/Home.php
EyePACS-AIROGS 0 3269 0 https://airogs.grand-challenge.org/data-and-challenge/
FIVES (Fundus Image VEssel Segmentation) 200 200 0 https://figshare.com/articles/figure/FIVES_A_Fundus_Image_Dataset_for_AI-based_Vessel_Segmentation/19688169/1
G1020 724 296 0 https://www.kaggle.com/datasets/arnavjain1/glaucoma-datasets
HRF (High Resolution Fundus) 15 15 0 https://www5.cs.fau.de/research/data/fundus-images/
JSIEC-1000 (Joint Shantou International Eye Center) 38 0 13 https://www.kaggle.com/datasets/linchundan/fundusimage1000
LES-AV 11 11 0 https://figshare.com/articles/dataset/LES-AV_dataset/11857698/1
OIA-ODIR-TRAIN 2932 197 18 https://www.kaggle.com/datasets/andrewmvd/ocular-disease-recognition-odir5k
OIA-ODIR-TEST-ONLINE 802 58 25 https://www.kaggle.com/datasets/andrewmvd/ocular-disease-recognition-odir5k
OIA-ODIR-TEST-OFFLINE 417 36 9 https://www.kaggle.com/datasets/andrewmvd/ocular-disease-recognition-odir5k
ORIGA-light 482 168 0 https://www.kaggle.com/datasets/sshikamaru/glaucoma-detection
PAPILA 333 87 68 https://doi.org/10.6084/m9.figshare.14798004.v1
REFUGE1-TRAIN (Retinal Fundus Glaucoma Challenge 1 Train) 360 40 0 https://refuge.grand-challenge.org/REFUGE2Download/
REFUGE1-VALIDATION (Retinal Fundus Glaucoma Challenge 1 Validation) 360 40 0 https://refuge.grand-challenge.org/REFUGE2Download/
sjchoi86-HRF 300 101 0 https://github.com/yiweichen04/retina_dataset
Total 7499 4817 133

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