CoChat

Predicting optical coherence tomography-derived diabetic macular edema grades from fundus photographs using deep learning

Open access
Varadarajan, Avinash V., Bavishi, Pinal, Ruamviboonsuk, Paisan, Chotcomwongse, Peranut, Venugopalan, Subhashini, Narayanaswamy, Arunachalam, Cuadros, Jorge, Kanai, Kuniyoshi, Bresnick, George, Tadarati, Mongkol, Silpa-archa, Sukhum, Limwattanayingyong, Jirawut, Nganthavee, Variya, Ledsam, Joseph R., Keane, Pearse A., Corrado, Greg S., Peng, Lily, Webster, Dale R. · 2020 · Nature Communications

The DOI matches its Crossref record; 37 of the 37 references it cites were found in a registry or an index.

Abstract

Center-involved diabetic macular edema (ci-DME) is a major cause of vision loss. Although the gold standard for diagnosis involves 3D imaging, 2D imaging by fundus photography is usually used in screening settings, resulting in high false-positive and false-negative calls. To address this, we train a deep learning model to predict ci-DME from fundus photographs, with an ROC–AUC of 0.89 (95% CI: 0.87–0.91), corresponding to 85% sensitivity at 80% specificity. In comparison, retinal specialists have similar sensitivities (82–85%), but only half the specificity (45–50%, p < 0.001). Our model can also detect the presence of intraretinal fluid (AUC: 0.81; 95% CI: 0.81–0.86) and subretinal fluid (AUC 0.88; 95% CI: 0.85–0.91). Using deep learning to make predictions via simple 2D images without sophisticated 3D-imaging equipment and with better than specialist performance, has broad relevance to many other applications in medical imaging.

Overview

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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

Integrity signals

Registry

Crossref · 10.1038/s41467-019-13922-8

Cites

37 of 37 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Illustration of our proposed approach for developing a ci-DME model. Ground truth for ci-DME were derived from a human grader analyzing the OCT for each case. In addition, subretinal fluid and intraretinal fluid presence grades were also collected. These ground truth labels and corresponding color fundus photos were used for model training. For clinical validation, the trained model takes in a new fundus photo and generates a predicted ci-DME grade, predicted subretinal fluid, and intraretinal fluid presence grades.
Fig. 1 Illustration of our proposed approach for developing a ci-DME model. Ground truth for ci-DME were derived from a human grader analyzing the OCT for each case. In addition, subretinal fluid and intraretinal fluid presence grades were also collected. These ground truth labels and corresponding color fundus photos were used for model training. For clinical validation, the trained model takes in a new fundus photo and generates a predicted ci-DME grade, predicted subretinal fluid, and intraretinal fluid presence grades.CC BY · PubMed Central
Fig. 2 Receiver operating characteristic curve of the model compared to retinal specialists. The retinal specialists’ grades for predicting ci-DME on the primary clinical validation set are shown as red dots. All methods (i.e., the model and retinal specialists) rendered their grades using monoscopic fundus images only. The ground truth for ci-DME was derived using OCT (center point thickness ≥ 250 μm).
Fig. 2 Receiver operating characteristic curve of the model compared to retinal specialists. The retinal specialists’ grades for predicting ci-DME on the primary clinical validation set are shown as red dots. All methods (i.e., the model and retinal specialists) rendered their grades using monoscopic fundus images only. The ground truth for ci-DME was derived using OCT (center point thickness ≥ 250 μm).CC BY · PubMed Central
Fig. 3 Effect of data size on prediction performance on the primary clinical validation set. A subsampled fraction of 1.0 indicates the entire training dataset. Model performance continues to increase with increased data, suggesting that the accuracy of predicting ci-DME, subretinal fluid, and intraretinal fluid presence will likely improve if the model is trained with more data. Error bars are 95% confidence interval using n = 2000 samples with replacement.
Fig. 3 Effect of data size on prediction performance on the primary clinical validation set. A subsampled fraction of 1.0 indicates the entire training dataset. Model performance continues to increase with increased data, suggesting that the accuracy of predicting ci-DME, subretinal fluid, and intraretinal fluid presence will likely improve if the model is trained with more data. Error bars are 95% confidence interval using n = 2000 samples with replacement.CC BY · PubMed Central
Fig. 4 Receiver operating characteristic curve of the model compared with eyepacs graders. The graders’ grades for predicting ci-DME on the secondary clinical validation set are shown as a red dots. All methods (i.e., the model and eyepacs graders) rendered their grades using monoscopic fundus images only. The ground truth for ci-DME was derived using OCT (central subfield thickness ≥ 300 μm).
Fig. 4 Receiver operating characteristic curve of the model compared with eyepacs graders. The graders’ grades for predicting ci-DME on the secondary clinical validation set are shown as a red dots. All methods (i.e., the model and eyepacs graders) rendered their grades using monoscopic fundus images only. The ground truth for ci-DME was derived using OCT (central subfield thickness ≥ 300 μm).CC BY · PubMed Central
Fig. 5 Model performance on crops around fovea and optic disk. The plot shows model performance in predicting ci-DME on the primary clinical validation set, as measured by AUC when cropped circular images are used to train and validate the model. The blue line indicates the performance when cropped circular images of different sizes (radius of multiples of disc diameter from 0.05 to 2.5) centered at the fovea are used; the green line indicates the corresponding performance when the crops are centered at the optic disc. (Inset) Image depicting some of the regions of different radii (0.05, 0.125, 0.25, 0.5, and 1 disc diameter) around the fovea and optic disc. Bottom panel: Fundus image, followed by a sample of the crops extracted by centering at the optic disc (green) and fovea (blue), with the extraction radius in multiples of disc diameter indicated in each crop. Error bars are 95% confidence interval using n = 2000 samples with replacement.
Fig. 5 Model performance on crops around fovea and optic disk. The plot shows model performance in predicting ci-DME on the primary clinical validation set, as measured by AUC when cropped circular images are used to train and validate the model. The blue line indicates the performance when cropped circular images of different sizes (radius of multiples of disc diameter from 0.05 to 2.5) centered at the fovea are used; the green line indicates the corresponding performance when the crops are centered at the optic disc. (Inset) Image depicting some of the regions of different radii (0.05, 0.125, 0.25, 0.5, and 1 disc diameter) around the fovea and optic disc. Bottom panel: Fundus image, followed by a sample of the crops extracted by centering at the optic disc (green) and fovea (blue), with the extraction radius in multiples of disc diameter indicated in each crop. Error bars are 95% confidence interval using n = 2000 samples with replacement.CC BY · PubMed Central

Verification

Claims the paper makes

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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?

Self-check

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

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

Citations

Cites

Show all 37 references

Research works printed but not linked: 1

  1. Chihara, L. M. & Hesterberg, T. C. Mathematical Statistics with Resampling and R (Wiley, 2011).

Cited by

See all 170

Metadata

Authors
Varadarajan, A. V., Bavishi, P., Ruamviboonsuk, P., Chotcomwongse, P., Venugopalan, S., Narayanaswamy, A., Cuadros, J., Kanai, K., Bresnick, G., Tadarati, M., Silpa-archa, S., Limwattanayingyong, J., Nganthavee, V., Ledsam, J. R., Keane, P. A., Corrado, G. S., Peng, L., & Webster, D. R.
Affiliations
  • Google (United States) (US)
  • Rajavithi Hospital (TH)
  • University of California, Berkeley (US)
  • Google DeepMind (United Kingdom) (GB)
  • Moorfields Eye Hospital NHS Foundation Trust (GB)
Corresponding author
Peng, Lily
Venue
Nature Communications
Published
2020-01-08
Volume
11
Issue
1
Pages
130-130
ISSN
2041-1723 (ISSN-L)
OpenAlex
W2896734945
Cited by
139
References
36
Citations by year
2026: 14 · 2025: 26 · 2024: 27 · 2023: 18 · 2022: 17 · 2021: 24 · 2020: 12
FWCI
8.64
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
  • eScholarship (California Digital Library) (submittedVersion, cc-by)
  • arXiv (Cornell University)
Funders
National Institute for Health and Care Research (CS-2014-14-023)
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 macular edema, fundus photography, deep learning, automated disease detection, intraretinal fluid, subretinal fluid, diabetic retinopathy screening, diagnostic accuracy
MeSH terms
Deep Learning, Aged, Diabetic Retinopathy, Female, Humans, Macular Edema, Male, Middle Aged, Mutation, Photography, Retina, Imaging, Three-Dimensional, Tomography, Optical Coherence

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Predicting optical coherence tomography-derived diabetic macular edema grades from fundus photographs using deep learning. https://app.cochat.ai/papers/10.1038/s41467-019-13922-8
The paper
Varadarajan, A. V., Bavishi, P., Ruamviboonsuk, P., Chotcomwongse, P., Venugopalan, S., Narayanaswamy, A., Cuadros, J., Kanai, K., Bresnick, G., Tadarati, M., Silpa-archa, S., Limwattanayingyong, J., Nganthavee, V., Ledsam, J. R., Keane, P. A., Corrado, G. S., Peng, L., & Webster, D. R. (2020). Predicting optical coherence tomography-derived diabetic macular edema grades from fundus photographs using deep learning. Nature Communications. https://doi.org/10.1038/s41467-019-13922-8

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.