Database Credentialed Access

mBRSET, a Mobile Brazilian Retinal Dataset

Luis Filipe Nakayama Lucas Zago Ribeiro David Restrepo Nathan Santos Barboza Raul Dias Fiterman Maria luiza Vieira Sousa Alexandre Durao Alves Pereira Caio Regatieri Fernando Korn Malerbi Rafael Andrade

Published: June 26, 2024. Version: 1.0

When using this resource, please cite: (show more options)
Nakayama, L. F., Zago Ribeiro, L., Restrepo, D., Santos Barboza, N., Dias Fiterman, R., Vieira Sousa, M. l., Pereira, A. D. A., Regatieri, C., Malerbi, F. K., & Andrade, R. (2024). mBRSET, a Mobile Brazilian Retinal Dataset (version 1.0). PhysioNet.

Please include the standard citation for PhysioNet: (show more options)
Goldberger, A., Amaral, L., Glass, L., Hausdorff, J., Ivanov, P. C., Mark, R., ... & Stanley, H. E. (2000). PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation [Online]. 101 (23), pp. e215–e220.


The Mobile Brazilian Retinal Dataset (mBRSET) presents a pioneering collection of retinal images captured via portable cameras, encompassing a diverse range of ethnic backgrounds from Itabuna, Bahia, Brazil. Comprising 5164 images from 1291 patients diagnosed with diabetes, this dataset is augmented with clinical and demographic metadata. Its significance lies in its provision as a resource for the development and validation of algorithms tailored for portable retinal cameras, which are increasingly being deployed, particularly in low and middle-income countries. By providing a dataset of retinal fundus photos captured via portable cameras, mBRSET offers an opportunity to develop and validate computer vision models for diabetic retinopathy. This resource has the potential to contribute to advancements in medical imaging and diagnostic technologies.


In ophthalmological practice, imaging serves as a pivotal tool for diagnosing and monitoring ocular conditions [1]. Retinal fundus photos specifically capture the ocular posterior segment, comprising the retina, optic disc, macula, and vessels, offering crucial information into ocular health during eye examinations conducted with specialized cameras.

The global prevalence of diabetes paints a concerning picture, with an estimated 540 million people affected worldwide, projected to escalate to 543 million by 2030 [2]. This epidemic afflicts approximately 10.5% of the adult population, with half unaware of their condition, and 75% residing in low- and middle-income countries [2]. Among the myriad complications associated with diabetes, diabetic retinopathy emerges as a primary concern, impacting about one-third of individuals diagnosed with diabetes. Left untreated, this condition can progress to sight-threatening levels in roughly 10% of affected patients [3].

The utilization of retinal fundus photos and telemedicine presents a promising avenue for remote diabetic retinopathy screening. Innovative approaches such as smartphone-based retinal imaging and teleophthalmology [4,5] hold potential to enhance coverage rates of diabetic retinopathy screening programs. Coupled with timely treatment, such strategies play a pivotal role in preventing visual impairment [4,6]. Traditionally, imaging has relied on standard tabletop fundus cameras, following the guidelines set forth in the Early Treatment Diabetic Retinopathy Study [7,8]. However, this method poses barriers to patient access due to its cumbersome nature, spatial requirements, complex acquisition process, and substantial capital investment, thereby limiting its widespread adoption [9,10].

The advent of smaller, portable devices heralds a more accessible and cost-effective screening process, particularly benefiting low-resource settings and underserved populations [6,9]. Additionally, artificial intelligence (AI) algorithms hold promise in improving medical care by facilitating screening, diagnosis, and monitoring, especially in resource-limited contexts. However, concerns persist regarding the fairness and accuracy of AI algorithms, stemming from non-representative data and biased models.

In low- and middle-income countries (LMICs), the shortage and irregular distribution of ophthalmologists relative to the population underscores the importance of affordable screening programs, which are increasingly being deployed using portable retinal cameras [5,11–13]. However, the existing datasets fail to adequately represent this new imaging modality [14].

mBRSET is the first publicly available diabetic retinopathy dataset comprised of images captured with handheld retina cameras in real-life, high-burden scenarios. This dataset encompasses individuals from diverse ethnic backgrounds, offering a comprehensive resource for addressing the challenges of diabetic retinopathy screening and management.


This study was approved by the institutional review board and included retinal fundus photos, clinical, and demographic data. Identifiable protected health information was removed from images and tabular data in accordance with the Health Insurance Portability and Accountability Act (HIPAA) and the Lei Geral de Proteção de Dados (LGPD) guidelines.

Data sources

We included data from patients who attended the Itabuna Diabetes Campaign, which took place in November 2022 in Itabuna, Bahia State, Northeast Brazil. Bahia State is a region with miscegenation. Nearly half of its population has European ancestry, around 40% trace their African ancestry, and 10% carry Native American ancestry [19]. This blend of ancestries creates a dynamic and multifaceted society, reflecting the unique fusion of cultures that defines Bahia. This annual event mobilizes many of the city’s inhabitants and involves diabetes awareness, counseling, screening, and treatment of diabetes complications [20]. After signing informed consent, participants answered a questionnaire for demographic and self-reported clinical characteristics and underwent ocular imaging.

Data collection

Images present in this dataset were collected using the Phelcom Eyer (Phelcom Technologies, Sao Carlos, Brazil), a portable, handheld, smartphone-based retinal fundus camera. Integrated into a Samsung Galaxy S10 running Android 11, the Eyer camera captures retinal images at a 45-degree angle, utilizing a high-resolution 12-megapixel sensor to produce images with dimensions of 1600x1600 pixels. The camera features an autofocus control with a range of -20 to +20 diopters. Trained non-medical personnel proficient in operating the Eyer camera captured the retinal photos under pharmacological mydriasis. Additionally, demographic information and relevant medical features were gathered through patient interviews.

Dataset preparation

The file identification was removed from all fundus photos, as well as sensitive data (e.g., patient name). Every image was reviewed to ensure the absence of protected health information in images. The images were exported from the Eyer cloud system in JPEG format, and no preprocessing techniques were performed. The image viewpoint can be macula-centered or optic disc-centered. The dataset does not include fluorescein angiogram photos and non-retinal images.


Two ophthalmologists, experts in retina and vitreous, meticulously labeled all the images according to predefined criteria set by the research group [21]. In cases of discordance in diabetic retinopathy grading, a third senior specialist provided adjudication. Kappa and weighted kappa statistics were employed to assess the agreement between graders, ensuring a robust evaluation of inter-rater reliability.

The following characteristics were labeled:

  • Quality control parameters: The evaluation of parameters encompasses the detection of artifacts, which encompass various issues such as image focus, illumination discrepancies, and the presence of dust. Additionally, each image is categorized based on its quality, thus determining whether it is deemed satisfactory or unsatisfactory for clinical assessment.
  • Diabetic retinopathy classification: Diabetic retinopathy and diabetic macular edema were classified using the International Clinic Diabetic Retinopathy (ICDR) grading [22].
    • 0 - Normal: No abnormalities
    • 1 - Mild non-proliferative DR: Microaneurysms only
    • 2 - Moderate non-proliferative DR: More than just microaneurysms but less than severe non-proliferative diabetic retinopathy
    • 3 - Severe non-proliferative DR: Any of the following: > 20 intra-retinal hemorrhages in each of 4 quadrants, definite venous beading in ≥2 quadrants, prominent intraretinal microvascular abnormalities in ≥1 quadrant, or no signs of proliferative retinopathy
    • 4 - Proliferative DR: One or more of the following: neovascularization and/or vitreous or preretinal hemorrhages and/or panphotocoagulation scars
    • Macular edema: Exudates or apparent thickening within one disc diameter from the fovea

Data Description

The dataset comprises of a comma-delimited metadata file containing labels (labels_mbrset.csv), along with an associated collection of images.

Metadata file

The metadata file (labels_mbrset.csv) includes the following columns:

  • patient: patient identifier.
  • age: patient age in years.
  • sex: 0 for female and 1 for male.
  • dm_time: diabetes diagnosis time in years.
  • insulin: self-referred use of insulin - 0 for no and 1 for yes.
  • insulin_time: insulin use time in years.
  • oraltreatment_dm: self-referred use of oral drugs for diabetes - 0 for no and 1 for yes
  • systemic_hypertension: self-referred diagnosis of systemic arterial hypertension 0 for no and 1 for yes
  • insurance: 0 for no and 1 for yes
  • educational_level: higher educational level
    • 1: Illiterate
    • 2: Incomplete Primary
    • 3: Complete Primary
    • 4: Incomplete Secondary
    • 5: Complete Secondary
    • 6: Incomplete Tertiary
    • 7: Complete Tertiary
    alcohol_consumption: self-referred regular alcohol consumption 0 for no and 1 for yes
  • smoking: self-referred active smoking - 0 for no and 1 for yes
  • obesity: self-referred obesity diagnosis - 0 for no and 1 for yes
  • vascular_disease: self-referred diabetes-related vascular disease - 0 for no and 1 for yes
  • acute_myocardial_infarction: self-referred previous acute myocardial infarction - 0 for no and 1 for yes
  • nephropathy: self-referred diabetes-related nephropathy - 0 for no and 1 for yes
  • neuropathy: self-referred diabetes-related neuropathy - 0 for no and 1 for yes
  • diabetic_foot: 0 for no and 1 for yes
  • file: image identifier
  • laterality: retinal fundus photo laterality
  • final_artifacts: presence of artifacts - yes or no
  • final_quality: final quality assessment - yes or no
  • final_icdr: ICDR score
    • 0 No retinopathy.
    • 1 Mild non-proliferative diabetic retinopathy.
    • 2 Moderate non-proliferative diabetic retinopathy.
    • 3 Severe non-proliferative diabetic retinopathy.
    • 4 Proliferative diabetic retinopathy and post-laser status.
  • final_edema: macular edema - yes or no

Patient level analysis

The dataset contains 5164 images from 1291 patients. The sex distribution is 451 (34.93%) male and 840 (65.06%) female patients. The average age is 61.44 years (SD 11.63). (Table 1)

Table 1: Patient's demographics characteristics.





sex, n (%)



840 (65.1)


451 (34.9)

insurance, n (%)



1180 (92.3)


99 (7.7)

educational_level, n (%)

Illiterate (1)


176 (13.8)

Incomplete Primary (2)

539 (42.2)

Complete Primary (3)

116 (9.1)

Incomplete Secondary (4)

120 (9.4)

Complete Secondary (5)

247 (19.3)

Incomplete Tertiary (6)

21 (1.6)

Complete Tertiary (7)

59 (4.6)

age, mean (SD)


61.4 (11.6)

The average duration of diabetes is 9.53 years (SD 8.64), with 1083 patients (84.61%) undergoing oral treatment and 266 patients (20.79%) receiving insulin therapy. (Table 2)

Table 2: Patient's clinical characteristics.





systemic_hypertension, n (%)



366 (28.6)


914 (71.4)

alcohol_consumption, n (%)



1094 (86.0)


178 (14.0)

smoking, n (%)



1188 (93.6)


81 (6.4)

obesity, n (%)



1170 (92.0)


102 (8.0)

vascular_disease, n (%)



1055 (82.9)


217 (17.1)

acute_myocardial_infarction, n (%)



1172 (92.3)


98 (7.7)

nephropathy, n (%)



1225 (96.4)


46 (3.6)

neuropathy, n (%)



1217 (95.7)


55 (4.3)

diabetic_foot, n (%)



1091 (86.3)


173 (13.7)

Image level analysis

In terms of the image findings, among the total dataset, 3759 images (76.79%) revealed no indications of diabetic retinopathy (DR). Mild non-proliferative DR was observed in 272 images (5.56%), while moderate non-proliferative DR was present in 570 images (11.64%). Additionally, 82 images (1.67%) of severe non-proliferative DR, and 212 images (4.33%) of proliferative DR. Furthermore, macular edema was identified in 427 images (8.69%). (Table 3)
Regarding inter-grader agreement, the Weighted Kappa for the ICDR score was 0.863, indicating near-perfect agreement, while for macular edema was 0.618, representing substantial agreement.

Table 3: Image level analysis.





laterality, n (%)



2584 (50.0)


2580 (50.0)

final_icdr, n (%)



3750 (76.8)


272 (5.6)


568 (11.6)


82 (1.7)


212 (4.3)

final_edema, n (%)



4472 (91.3)


427 (8.7)

referable_dr, n (%)



4167 (80.7)


997 (19.3)

Regarding the quality distribution, 4283 (82.7%) of the images have some artifacts, and 4833 (94.3%) have good quality that enables image assessment.

Usage Notes

This is the first mobile camera retinal fungus dataset, and future releases may increase the dataset size and provide race information. Our objective is to reduce the under-represented countries in the ophthalmology dataset pool and enable validation studies using this modality.

All the codes used in this paper for the dataset setup, data analysis, and experiments are found in a GitHub repository at [23]. Best practice guidelines should be followed when analyzing the data, and we incentivize sharing codes and models to promote reproducibility.

Release Notes

v1.0: First public release. In future releases, we plan to expand the dataset with more images and patients from screening programs and add self-declared races.


This study was approved by the INSTITUTO DE ENSINO SUPERIOR PRESIDENTE TANCREDO DE ALMEIDA NEVES (IPTAN) institutional review board (CAAE 64219922.3.0000.9667). All participants provided written informed consent, and the sharing of data was authorized by both patients and institutions.

Conflicts of Interest

The authors declare no conflicts of interest to declare.


  1. Nakayama LF, Matos J, Quion J, Novaes F, Mitchell WG, Mwavu R, et al. Unmasking biases and navigating pitfalls in the ophthalmic Artificial Intelligence lifecycle: A review. 2023. doi:10.48550/ARXIV.2310.04997
  2. Sun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB, et al. IDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. Diabetes Res Clin Pract. 2022;183: 109119. doi:10.1016/j.diabres.2021.109119
  3. Sivaprasad S, Pearce E. The unmet need for better risk stratification of non-proliferative diabetic retinopathy. Diabet Med. 2019;36: 424–433. doi:10.1111/dme.13868
  4. Lim G, Bellemo V, Xie Y, Lee XQ, Yip MYT, Ting DSW. Different fundus imaging modalities and technical factors in AI screening for diabetic retinopathy: a review. Eye Vis (Lond). 2020;7: 21. doi:10.1186/s40662-020-00182-7
  5. Sengupta S, Sindal MD, Besirli CG, Upadhyaya S, Venkatesh R, Niziol LM, et al. Screening for vision-threatening diabetic retinopathy in South India: comparing portable non-mydriatic and standard fundus cameras and clinical exam. Eye . 2018;32: 375–383. doi:10.1038/eye.2017.199
  6. Salongcay RP, Aquino LAC, Salva CMG, Saunar AV, Alog GP, Sun JK, et al. Comparison of Handheld Retinal Imaging with ETDRS 7-Standard Field Photography for Diabetic Retinopathy and Diabetic Macular Edema. Ophthalmol Retina. 2022;6: 548–556. doi:10.1016/j.oret.2022.03.002
  7. Vujosevic S, Benetti E, Massignan F, Pilotto E, Varano M, Cavarzeran F, et al. Screening for diabetic retinopathy: 1 and 3 nonmydriatic 45-degree digital fundus photographs vs 7 standard early treatment diabetic retinopathy study fields. Am J Ophthalmol. 2009;148: 111–118. doi:10.1016/j.ajo.2009.02.031
  8. Vujosevic S, Aldington SJ, Silva P, Hernández C, Scanlon P, Peto T, et al. Screening for diabetic retinopathy: new perspectives and challenges. Lancet Diabetes Endocrinol. 2020;8: 337–347. doi:10.1016/S2213-8587(19)30411-5
  9. Piyasena MMPN, Yip JLY, MacLeod D, Kim M, Gudlavalleti VSM. Diagnostic test accuracy of diabetic retinopathy screening by physician graders using a hand-held non-mydriatic retinal camera at a tertiary level medical clinic. BMC Ophthalmol. 2019;19: 89. doi:10.1186/s12886-019-1092-3
  10. Midena E, Zennaro L, Lapo C, Torresin T, Midena G, Pilotto E, et al. Handheld Fundus Camera for Diabetic Retinopathy Screening: A Comparison Study with Table-Top Fundus Camera in Real-Life Setting. J Clin Med Res. 2022;11. doi:10.3390/jcm11092352
  11. Malerbi FK, Andrade RE. Real-World diabetic retinopathy screening with a handheld fundus camera in a high-burden setting. Acta Ophthalmol. 2022;100: e1771. doi:10.1111/aos.15170
  12. Jin K, Lu H, Su Z, Cheng C, Ye J, Qian D. Telemedicine screening of retinal diseases with a handheld portable non-mydriatic fundus camera. BMC Ophthalmol. 2017;17: 89. doi:10.1186/s12886-017-0484-5
  13. Caceres J, Zhang Y, Boe L, Zhou Y, Besirli C, Paulus YM, et al. Diabetic Retinopathy Screening Using a Portable Retinal Camera in Vanuatu. Clin Ophthalmol. 2023;17: 2919–2927. doi:10.2147/OPTH.S410425
  14. Khan SM, Liu X, Nath S, Korot E, Faes L, Wagner SK, et al. A global review of publicly available datasets for ophthalmological imaging: barriers to access, usability, and generalisability. Lancet Digit Health. 2021;3: e51–e66. doi:10.1016/s2589-7500(20)30240-5
  15. Nakayama LF, Goncalves M, Zago Ribeiro L, Santos H, Ferraz D, Malerbi F, et al. A Brazilian Multilabel Ophthalmological Dataset (BRSET). PhysioNet; 2023. doi:10.13026/XCXW-8198
  16. Cuadros J, Bresnick G. EyePACS: an adaptable telemedicine system for diabetic retinopathy screening. J Diabetes Sci Technol. 2009;3: 509–516. doi:10.1177/193229680900300315
  17. Decencièreb E, Zhang X, Cazuguel G, Laÿ B, Cochener B, Trone C, et al. FEEDBACK ON A PUBLICLY DISTRIBUTED IMAGE DATABASE: THE MESSIDOR DATABASE. doi:10.5566/ias.1155
  18. Pachade S, Porwal P, Thulkar D, Kokare M, Deshmukh G, Sahasrabuddhe V, et al. Retinal Fundus Multi-Disease Image Dataset (RFMiD): A Dataset for Multi-Disease Detection Research. Brown Univ Dig Addict Theory Appl. 2021;6: 14. doi:10.3390/data6020014
  19. Souza AM de, Resende SS, Sousa TN de, Brito CFA de. A systematic scoping review of the genetic ancestry of the Brazilian population. Genet Mol Biol. 2019;42: 495–508. doi:10.1590/1678-4685-GMB-2018-0076
  20. Malerbi FK, Andrade RE, Morales PH, Stuchi JA, Lencione D, de Paulo JV, et al. Diabetic Retinopathy Screening Using Artificial Intelligence and Handheld Smartphone-Based Retinal Camera. J Diabetes Sci Technol. 2021; 1932296820985567. doi:10.1177/1932296820985567
  21. Nakayama LF, Gonçalves MB, Ribeiro LZ, Malerbi FK, Regatieri CVS. Diabetic Retinopathy Labeling Protocol for the Brazilian Multilabel Ophthalmological Dataset. 2023. doi:10.31219/
  22. Wilkinson CP, Ferris FL 3rd, Klein RE, Lee PP, Agardh CD, Davis M, et al. Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales. Ophthalmology. 2003;110: 1677–1682. doi:10.1016/S0161-6420(03)00475-5
  23. Nakayama L. mBRSET: Repository of the Mobile Brazilian Retinal Dataset (mBRSET). Github; Available:

Parent Projects
mBRSET, a Mobile Brazilian Retinal Dataset was derived from: Please cite them when using this project.

Access Policy:
Only credentialed users who sign the DUA can access the files.

License (for files):
PhysioNet Credentialed Health Data License 1.5.0

Data Use Agreement:
PhysioNet Credentialed Health Data Use Agreement 1.5.0

Required training:
CITI Data or Specimens Only Research

Corresponding Author
You must be logged in to view the contact information.