bumbledeep/rfmid

Dataset Card for Dataset Name

All the images of the dataset come from this kaggle dataset. Some minor modifications have been made to the metadata. All credit goes to the original authors and the contributor on Kaggle.

Dataset Details

Dataset Description

Retinal Fundus Multi-disease Image Dataset (RFMiD) consists of 3200 fundus images captured using three different fundus cameras with 46 conditions annotated through adjudicated consensus of two senior retinal experts. To the best of our knowledge, our dataset, RFMiD, is the only publicly available dataset that constitutes such a wide variety of diseases that appear in routine clinical settings. This dataset will enable the development of generalizable models for retinal screening.

• Funded by: This work was supported by Center of Excellence in Signal and Image Processing, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded, India.
• Curated by: Pachade et al. (2021)
• Shared by: Larxel
• License: CC-BY 4.0

Dataset Sources

• Original Repository: https://riadd.grand-challenge.org/download-all-classes/
• Paper [optional]: Pachade et al. (2021).

Uses

Direct Use

• Eye disease classification (multilabel -> using label or single label -> using Disease_Risk).
• Feature extraction (unsupervised or self supervised learning).

Out-of-Scope Use

[More Information Needed]

Dataset Structure

The RFMiD is a new publicly available retinal images dataset consisting of 3200 images along with the expert annotations divided into two categories, as follows: Screening of retinal images into normal and abnormal (comprising of 45 different types of diseases/pathologies) categories. Classification of retinal images into 45 different categories. The dataset is split into 3 subsets: training 60% (1920 images), evaluation 20% (640 images), and test 20% (640 images) sets. The disease wise stratification on average in training, evaluation and test set is 60 ± 7 %, 20 ± 7%, and 20 ± 5%, respectively.

For more detailed information, see Pachade et al. (2021).

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

Data Collection and Processing

he retinal images were acquired using three different digital fundus cameras, TOPCON 3D OCT-2000, Kowa VX-10𝛼, and TOPCON TRC-NW300, all of them centered either on the macula or optic disc. These images are collected from subjects visiting an eye clinic due to a concern for their eye health.

• Pretreatment of Samples: Before image acquisition, pupils of most of the subjects were dilated with one drop of tropicamide at 0.5% concentration. The fundus images were captured with position and orientation of the patient sitting upright with 39 mm (Kowa VX–10) and 40.7 mm (TOPCON 3D OCT-2000 and TOPCON TRC-NW300) distance between lenses and examined eye using non-invasive fundus camera.
• Fundus Camera Specifications: Regular retinal fundus images were acquired using three different digital fundus cameras. Details of camera model, hardware used, field of view (FOV), resolution, and number of images included in the dataset are given in Table 2.
• Data Quality: The dataset is formed by extracting 3200 images from the thousands of examinations done during the period 2009–2020. Both high-quality and low-quality images are selected to make the dataset challenging.

For more detailed information, see Pachade et al. (2021).

Annotation process

Initially, all the images were labeled by two ophthalmologists independently. The reference standard for presence of different diseases was assigned based on the comprehensive evaluation of the subjects clinical records and visual fields. If a fundus image shows presence of more than one disease, then multiple labels are assigned to the single image. After the ophthalmologists completed initial labeling of fundus photographs, the leader of the project team checked and confirmed or corrected through consultation from both the ophthalmologists when difference in diagnostic assessments was observed to get adjudicated consensus for the labels.

For more detailed information, see Pachade et al. (2021).

Personal and Sensitive Information

Informed consent was obtained from all subjects involved in the study.

Bias, Risks, and Limitations

[More Information Needed]

Recommendations

Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.

Citation

Samiksha Pachade, Prasanna Porwal, Dhanshree Thulkar, Manesh Kokare, Girish Deshmukh, Vivek Sahasrabuddhe, Luca Giancardo, Gwenolé Quellec, and Fabrice Mériaudeau, 2021. Retinal Fundus Multi-Disease Image Dataset (RFMiD): A Dataset for Multi-Disease Detection Research. Data, 6(2), p.14. Available (Open Access): https://www.mdpi.com/2306-5729/6/2/14

Glossary

0. Diabetic retinopathy (DR).
1. Age-related macular degeneration (ARMD).
2. Media Haze (MH).
3. Drusens (DN).
4. Myopia (MYA).
5. Branch retinal vein occlusion (BRVO).
6. Tessellation (TSLN).
7. Epiretinal membrane (ERM).
8. Laser scars (LS).
9. Macular scar (MS).
10. Central serous retinopathy (CSR).
11. Optic disc cupping (ODC).
12. Central retinal vein occlusion (CRVO).
13. Tortuous vessels (TV).
14. Asteroid hyalosis (AH).
15. Optic disc pallor (ODP).
16. Optic disc edema (ODE).
17. Optociliary shunt (ST).
18. Anterior ischemic optic neuropathy (AION).
19. Parafoveal telangiectasia (PT).
20. Retinal traction (RT).
21. Retinitis (RS).
22. Chorioretinitis (CRS).
23. Exudation (EDN).
24. Retinal pigment epithelium changes (RPEC).
25. Macular hole (MHL).
26. Retinitis pigmentosa (RP).
27. Cotton-wool spots (CWS).
28. Coloboma (CB).
29. Optic disc pit maculopathy (ODPM).
30. Preretinal hemorrhage (PRH).
31. Myelinated nerve fibers (MNF).
32. Hemorrhagic retinopathy (HR).
33. Central retinal artery occlusion (CRAO).
34. Tilted disc (TD).
35. Cystoid macular edema (CME).
36. Post-traumatic choroidal rupture (PTCR).
37. Choroidal folds (CF).
38. Vitreous hemorrhage (VH).
39. Macroaneurysm (MCA).
40. Vasculitis (VS).
41. Branch retinal artery occlusion (BRAO).
42. Plaque (PLQ).
43. hemorrhagic pigment epithelial detachment (HPED).
44. Collateral (CL).

For more detailed information, see Pachade et al. (2021).

Dataset Card Authors

bumbledeep