CoChat
Sign Up
Cited by 620Retinal Imaging and AnalysisRetinal Diseases and TreatmentsOphthalmology and Visual Health Research

A deep learning system for detecting diabetic retinopathy across the disease spectrum

Open access
Dai, Ling, Wu, Liang, Li, Huating, Cai, Chun, Wu, Qiang, Kong, Hongyu, Liu, Ruhan, Wang, Xiangning +12 more·2021·Nature Communications
DOI 10.1038/s41467-021-23458-5PMC PMC8163820
Open this paper in CoChat — free accountOpen PDF

Abstract

Retinal screening contributes to early detection of diabetic retinopathy and timely treatment. To facilitate the screening process, we develop a deep learning system, named DeepDR, that can detect early-to-late stages of diabetic retinopathy. DeepDR is trained for real-time image quality assessment, lesion detection and grading using 466,247 fundus images from 121,342 patients with diabetes. Evaluation is performed on a local dataset with 200,136 fundus images from 52,004 patients and three external datasets with a total of 209,322 images. The area under the receiver operating characteristic curves for detecting microaneurysms, cotton-wool spots, hard exudates and hemorrhages are 0.901, 0.941, 0.954 and 0.967, respectively. The grading of diabetic retinopathy as mild, moderate, severe and proliferative achieves area under the curves of 0.943, 0.955, 0.960 and 0.972, respectively. In external validations, the area under the curves for grading range from 0.916 to 0.970, which further supports the system is efficient for diabetic retinopathy grading.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41467-021-23458-5

Cites

10 of 10 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Screening of DR using retinal images was studied well in multi-ethnic cohorts, with comparable accuracy compared with professional graders.

    “In the screening and evaluation of diabetic complications, the screening of DR using retinal images was studied well in multi-ethnic cohorts, with comparable accuracy compared with professional graders (discussed in detail in the conclusion and perspective).103,104,105,106,107,108,109”

    From Artificial intelligence in diabetes management: Advancements, opportunities, and challenges (2023) · cites this paper as [103]

    The Results (External validation; page unavailable) report high DR-grading accuracy from retinal images in Chinese cohorts and the US EyePACS dataset. But the Discussion says DeepDR was developed using a single-ethnic cohort and that further multiethnic validation is needed. The Methods describe grading by certified ophthalmologists as the reference process, not a direct finding that DeepDR’s accuracy was comparable to theirs. (LLM, conf=0.86)

    Experimental — accuracy not yet measured

  • Dai et al. reported that DeepDR (CNN) using retinal images for DR screening achieved AUCs of 0.943, 0.955, 0.960, and 0.972 for grading DR as mild, moderate, severe, and proliferative, respectively, in internal validation, and AUCs from 0.916 to 0.970 in external validations.

    “the grading of DR as mild, moderate, severe, and proliferative achieves AUCs of 0.943, 0.955, 0.960, and 0.972, respectively (internal validation); in external validations, AUCs for grading range from 0.916 to 0.970”

    From Artificial intelligence in diabetes management: Advancements, opportunities, and challenges (2023) · cites this paper as Dai et al.103

    Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

    Experimental — accuracy not yet measured

  • A deep learning-based AI system developed by Dai et al. could predict the progression of diabetic retinopathy by analyzing subtle changes in retinal microvasculature not easily detectable through traditional screening methods.

    “A study carried out by Dai et al. demonstrated that a deep learning-based AI system could predict the progression of diabetic retinopathy by analyzing subtle changes in retinal microvasculature that are not easily detectable through traditional screening methods (92,93).”

    From Generative artificial intelligence in predictive analysis of diabetes and its complications: a narrative review (2025) · cites this paper as [93]

    The Introduction and Results describe DeepDR as a system for detecting retinal lesions and grading diabetic retinopathy. The Discussion says its lesion-aware sub-network has the potential to capture tiny changes during follow-up, but also says further studies are needed to evaluate detection and prediction of progression. The paper does not demonstrate progression prediction or establish that the changes are undetectable by traditional screening (page numbers not provided). (LLM, conf=0.89)

    Experimental — accuracy not yet measured

  • The AI model was able to predict the likelihood of disease progression, improve image collection quality, provide clinical reference, and facilitate diabetic retinopathy screening with an accuracy rate that rivaled expert ophthalmologists.

    “The AI model was able to predict the likelihood of disease progression, improve image collection quality, provide clinical reference, and facilitate diabetic retinopathy screening with an accuracy rate that rivaled expert ophthalmologists, offering the potential for earlier intervention.”

    From Generative artificial intelligence in predictive analysis of diabetes and its complications: a narrative review (2025) · cites this paper as [93]

    The Discussion (page not provided) says DeepDR can improve image collection quality, provide clinical reference, and facilitate diabetic retinopathy screening; the Results report high grading performance. But the Discussion says further studies are needed to evaluate prediction of disease progression, so the claim that the model could already predict it is unsupported. The passages found do not establish that its accuracy rivaled expert ophthalmologists. (LLM, conf=0.90)

    Experimental — accuracy not yet measured

  • Dai et al. developed an automated, interpretable, and validated system (DeepDR) that performs real-time image quality feedback, retinal lesion detection, and early- to late-stage DR grading with high sensitivity and specificity.

    “To overcome this issue, Dai et al. developed an automated, interpretable, and validated system (named DeepDR) that performs real-time image quality feedback, retinal lesion detection, and early- to late-stage DR grading with high sensitivity and specificity.103”

    From Artificial intelligence in diabetes management: Advancements, opportunities, and challenges (2023) · cites this paper as [103]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 44

  1. [2]American Diabetes Association. 11. Microvascular complications and foot care: standards of medical care in diabetes-2020 Diabetes Care 2020 43 S135 S151 10.2337/dc20-S011 31862754 · DOI · Paper page
  2. [3]Wang LZ Availability and variability in guidelines on diabetic retinopathy screening in Asian countries Br. J. Ophthalmol. 2017 101 1352 1360 10.1136/bjophthalmol-2016-310002 28292772 · DOI
  3. [4]Liew G Michaelides M Bunce C A comparison of the causes of blindness certifications in England and Wales in working age adults (16–64 years), 1999–2000 with 2009–2010 BMJ Open 2014 4 e004015 10.1136/bmjopen-2013-004015 24525390 PMC3927710 · DOI
  4. [5]Kristinsson JK Hauksdottir H Stefansson E Jónasson F Gíslason I Active prevention in diabetic eye disease: a 4-year follow-up Acta Ophthalmol. Scand. 1997 75 249 254 10.1111/j.1600-0420.1997.tb00766.x 9253967 · DOI
  5. [7]Kung K Prevalence of complications among Chinese diabetic patients in urban primary care clinics: a cross-sectional study BMC Fam. Pract. 2014 15 8 10.1186/1471-2296-15-8 24410810 PMC3913315 · DOI
  6. [8]Hu Y Prevalence and risk factors of diabetes and diabetic retinopathy in Liaoning province, China: a population-based cross-sectional study PLoS ONE 2015 10 e0121477 10.1371/journal.pone.0121477 25785633 PMC4364908 · DOI
  7. [9]Pang C Determination of diabetic retinopathy prevalence and associated risk factors in Chinese diabetic and pre-diabetic subjects: Shanghai diabetic complications study Diabetes Metab. Res. Rev. 2012 28 276 283 10.1002/dmrr.1307 22139892 · DOI
  8. [10]Lian JX Systematic screening for diabetic retinopathy (DR) in Hong Kong: prevalence of DR and visual impairment among diabetic population Br. J. Ophthalmol. 2015 100 151 155 10.1136/bjophthalmol-2015-307382 26271268 · DOI
  9. [11]Ting DSW Cheung GCM Wong TY Diabetic retinopathy: global prevalence, major risk factors, screening practices and public health challenges: a review Clin. Exp. Ophthalmol. 2016 44 260 277 10.1111/ceo.12696 26716602 · DOI · Paper page
  10. [12]Hu C Jia W Diabetes in China: epidemiology and genetic risk factors and their clinical utility in personalized medication Diabetes 2018 67 3 11 10.2337/dbi17-0013 29263166 · DOI
  11. [13]Li L Fu H China’s health care system reform: progress and prospects Int. J. Health Plann. Manage. 2017 32 240 253 10.1002/hpm.2424 28612498 · DOI
  12. [15]Silver D Mastering the game of Go without human knowledge Nature 2017 550 354 359 10.1038/nature24270 29052630 · DOI · Paper page
  13. [17]Shen D Wu G Suk H-I Deep learning in medical image analysis Annu. Rev. Biomed. Eng. 2017 19 221 248 10.1146/annurev-bioeng-071516-044442 28301734 PMC5479722 · DOI · Paper page
  14. [20]Gargeya R Leng T Automated identification of diabetic retinopathy using deep learning Ophthalmology 2017 124 962 969 10.1016/j.ophtha.2017.02.008 28359545 · DOI
  15. [21]Gulshan V Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs JAMA 2016 316 2402 2410 10.1001/jama.2016.17216 27898976 · DOI · Paper page
  16. [22]Ting DSW Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes JAMA 2017 318 2211 2223 10.1001/jama.2017.18152 29234807 PMC5820739 · DOI · Paper page
  17. [23]van der Heijden AA Validation of automated screening for referable diabetic retinopathy with the IDx-DR device in the Hoorn Diabetes Care System Acta Ophthalmol. 2018 96 63 68 10.1111/aos.13613 29178249 PMC5814834 · DOI · Paper page
  18. [24]Li Z An automated grading system for detection of vision-threatening referable diabetic retinopathy on the basis of color fundus photographs Diabetes Care 2018 41 2509 2516 10.2337/dc18-0147 30275284 · DOI · Paper page
  19. [25]Liu Y Glycemic exposure and blood pressure influencing progression and remission of diabetic retinopathy: a longitudinal cohort study in GoDARTS Diabetes Care 2013 36 3979 3984 10.2337/dc12-2392 24170761 PMC3836116 · DOI
  20. [26]Kendall, A., Gal, Y. & Cipolla, R. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (IEEE, 2018).
  21. [27]Wang, X., Ju, L., Zhao, X. & Ge, Z. Retinal abnormalities recognition using regional multitask learning. In International Conference on Medical Image Computing and Computer-Assisted Intervention (eds Shen, D. et al.) 30–38 (Springer, 2019).
  22. [28]Playout C Duval R Cheriet F A novel weakly supervised multitask architecture for retinal lesions segmentation on fundus images IEEE Trans. Med. Imaging 2019 38 2434 2444 10.1109/TMI.2019.2906319 30908197 · DOI
  23. [29]Yosinski, J., Clune, J., Bengio, Y. & Lipson, H. How transferable are features in deep neural networks? In International Conference on Neural Information Processing Systems 3320–3328 (NIPS, 2014).
  24. [30]Scanlon PH Comparison of two reference standards in validating two field mydriatic digital photography as a method of screening for diabetic retinopathy Br. J. Ophthalmol. 2003 87 1258 1263 10.1136/bjo.87.10.1258 14507762 PMC1920793 · DOI
  25. [34]Bakri SJ Kaiser PK Posterior subtenon triamcinolone acetonide for refractory diabetic macular edema Am. J. Ophthalmol. 2005 139 290 294 10.1016/j.ajo.2004.09.038 15733990 · DOI
  26. [36]Rodriguez-Gutierrez R Gonzalez-Gonzalez JG Zuñiga-Hernandez JA McCoy RG Benefits and harms of intensive glycemic control in patients with type 2 diabetes BMJ 2019 367 l5887 10.1136/bmj.l5887 31690574 · DOI · Paper page
  27. [37]Qaseem A Hemoglobin A1c targets for glycemic control with pharmacologic therapy for nonpregnant adults with type 2 diabetes mellitus: a guidance statement update from the American College of Physicians Ann. Intern. Med. 2018 168 569 576 10.7326/M17-0939 29507945 · DOI
  28. [39]Flaxel CJ Diabetic retinopathy preferred practice pattern® Ophthalmology 2020 127 P66 P145 10.1016/j.ophtha.2019.09.025 31757498 · DOI
  29. [40]Eftekhari N Pourreza H-R Masoudi M Ghiasi-Shirazi K Saeedi E Microaneurysm detection in fundus images using a two-step convolutional neural network Biomed. Eng. Online 2019 18 67 10.1186/s12938-019-0675-9 31142335 PMC6542103 · DOI
  30. [43]Keech AC Effect of fenofibrate on the need for laser treatment for diabetic retinopathy (FIELD study): a randomised controlled trial Lancet 2007 370 1687 1697 10.1016/S0140-6736(07)61607-9 17988728 · DOI · Paper page
  31. [44]Ying G-s Association between antiplatelet or anticoagulant drugs and retinal or subretinal hemorrhage in the comparison of age-related macular degeneration treatments trials Ophthalmology 2016 123 352 360 10.1016/j.ophtha.2015.09.046 26545320 PMC4724480 · DOI
  32. [45]Ribeiro ML Nunes SG Cunha-Vaz JG Microaneurysm turnover at the macula predicts risk of development of clinically significant macular edema in persons with mild nonproliferative diabetic retinopathy Diabetes Care 2013 36 1254 1259 10.2337/dc12-1491 23204247 PMC3631829 · DOI
  33. [46]Hove MN Kristensen JK Lauritzen T Bek T Quantitative analysis of retinopathy in type 2 diabetes: identification of prognostic parameters for developing visual loss secondary to diabetic maculopathy Acta Ophthalmol. Scand. 2004 82 679 685 10.1111/j.1600-0420.2004.00364.x 15606463 · DOI
  34. [47]Klein R Klein BE Moss SE How many steps of progression of diabetic retinopathy are meaningful? The Wisconsin Epidemiologic Study of Diabetic Retinopathy Arch. Ophthalmol. 2001 119 547 553 10.1001/archopht.119.4.547 11296020 · DOI
  35. [48]Papernot, N. et al. Practical black-box attacks against machine learning. In Proc. 2017 ACM on Asia Conference on Computer and Communications Security 506–519 (ACM, 2017).
  36. [50]Abràmoff MD Lavin PT Birch M Shah N Folk JC Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices npj Digital Med. 2018 1 1 8 10.1038/s41746-018-0040-6 PMC6550188 31304320 · DOI · Paper page
  37. [51]Fleming AD Philip S Goatman KA Olson JA Sharp PF Automated assessment of diabetic retinal image quality based on clarity and field definition Invest. Ophthalmol. Vis. Sci. 2006 47 1120 1125 10.1167/iovs.05-1155 16505050 · DOI
  38. [52]Boucher MC Gresset JA Angioi K Olivier S Effectiveness and safety of screening for diabetic retinopathy with two nonmydriatic digital images compared with the seven standard stereoscopic photographic fields Can. J. Ophthalmol. 2003 38 557 568 10.1016/S0008-4182(03)80109-6 14740797 · DOI
  39. [53]Wilkinson C Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales Ophthalmology 2003 110 1677 1682 10.1016/S0161-6420(03)00475-5 13129861 · DOI
  40. [54]Group ETDRSR. Treatment techniques and clinical guidelines for photocoagulation of diabetic macular edema: Early Treatment Diabetic Retinopathy Study report number 2 Ophthalmology 1987 94 761 774 10.1016/S0161-6420(87)33527-4 3658348 · DOI · Paper page
  41. [56]Shen Y Domain-invariant interpretable fundus image quality assessment Med. Image Anal. 2020 61 101654 10.1016/j.media.2020.101654 32066065 · DOI
  42. [57]He, K., Gkioxari, G., Dollár, P. & Girshick, R. Mask R-CNN. In Proc. IEEE International Conference on Computer Vision 2980–2988 (IEEE, 2017).
  43. [58]Ruder S. An overview of multi-task learning in deep neural networks. Preprint at https://arxiv.org/abs/1706.05098 (2017). · arXiv
  44. [59]McClish DK Analyzing a portion of the ROC curve Med. Decis. Making 1989 9 190 195 10.1177/0272989X8900900307 2668680 · DOI

Cited by

Sorted by citations

Authors
Dai, L., Wu, L., Li, H., Cai, C., Wu, Q., Kong, H., Liu, R., Wang, X., Hou, X., Liu, Y., Long, X., Wen, Y., Lu, L., Shen, Y., Chen, Y., Shen, D., Yang, X., Zou, H., Sheng, B., & Jia, W.
Affiliations
  • Shanghai Jiao Tong University (CN)
  • ShanghaiTech University (CN)
Corresponding author
  • Zou, Haidong
  • Sheng, Bin
  • Jia, Weiping
ORCID
Venue
Nature Communications
Publisher
Springer Science and Business Media LLC
Published
2021-05-28
Volume
12
Issue
1
Pages
3242-3242
ISSN
2041-1723 (ISSN-L)
DOI
10.1038/s41467-021-23458-5
PMC
PMC8163820
OpenAlex
W3164440025
Cited by
620
References
10
Citations by year
2026: 84 · 2025: 166 · 2024: 181 · 2023: 113 · 2022: 68 · 2021: 7 · 2017: 1
FWCI
48.74
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Natural Science Foundation of China (81561128016, 61872241)
  • Science and Technology Commission of Shanghai Municipality (17411952600, 16DZ0501100, 18410750700)
  • Medical Research Council
  • National Health and Medical Research Council (81561128016)
Topics
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Retinal Imaging and Analysis
  • Health Sciences › Medicine › Ophthalmology › Retinal Diseases and Treatments
  • Health Sciences › Medicine › Ophthalmology › Ophthalmology and Visual Health Research
Keywords
diabetic retinopathy, retinal screening, fundus images, disease grading, lesion detection, microaneurysms, cotton wool spots, hard exudates, retinal hemorrhage, image quality assessment, external validation
MeSH terms
Deep Learning, Aged, Diabetes Mellitus, Type 2, Diabetic Retinopathy, Fundus Oculi, Humans, Image Interpretation, Computer-Assisted, ROC Curve, Severity of Illness Index, Datasets as Topic

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review