Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 42 of the 42 references it cites were found in a registry or an index.

Deep learning algorithm predicts diabetic retinopathy progression in individual patients

Open access
Arcadu, Filippo, Benmansour, Fethallah, Maunz, Andreas, Willis, Jeff, Haskova, Zdenka, Prunotto, Marco · 2019 · npj Digital Medicine

Abstract

The global burden of diabetic retinopathy (DR) continues to worsen and DR remains a leading cause of vision loss worldwide. Here, we describe an algorithm to predict DR progression by means of deep learning (DL), using as input color fundus photographs (CFPs) acquired at a single visit from a patient with DR. The proposed DL models were designed to predict future DR progression, defined as 2-step worsening on the Early Treatment Diabetic Retinopathy Diabetic Retinopathy Severity Scale, and were trained against DR severity scores assessed after 6, 12, and 24 months from the baseline visit by masked, well-trained, human reading center graders. The performance of one of these models (prediction at month 12) resulted in an area under the curve equal to 0.79. Interestingly, our results highlight the importance of the predictive signal located in the peripheral retinal fields, not routinely collected for DR assessments, and the importance of microvascular abnormalities. Our findings show the feasibility of predicting future DR progression by leveraging CFPs of a patient acquired at a single visit. Upon further development on larger and more diverse datasets, such an algorithm could enable early diagnosis and referral to a retina specialist for more frequent monitoring and even consideration of early intervention. Moreover, it could also improve patient recruitment for clinical trials targeting DR.

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/s41746-019-0172-3

Cites

42 of 42 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 An overview of retinal imaging features analyzed to assess diabetic retinopathy (DR) severity and a schematic of the study design. a Example of fovea-centered color fundus photographs (CFPs) of a patient without DR (left) and a patient with signs of DR (right). In the CFP of the patient with signs of DR (right), one example each of hemorrhage, exudate, and a microaneurysm are highlighted. Both examples have been selected from the Kaggle DR dataset.47 b Schematic of the Diabetic Retinopathy Severity Scale (DRSS) established by the Early Treatment Diabetic Retinopathy Study (ETDRS) group to measure DR worsening over time. c Schematic of the two-phase modeling to detect two-step or more DRSS worsening over time. In phase I, field-specific Inception-v3 deep convolutional neural networks (DCNNs) called “field-specific DCNNs” or “pillars” are trained by means of transfer learning to predict whether the patient will progress two ETDRS DRSS steps. In phase II, the probabilities independently generated by the field-specific DCNNs are aggregated by means of random forest
Fig. 1 An overview of retinal imaging features analyzed to assess diabetic retinopathy (DR) severity and a schematic of the study design. a Example of fovea-centered color fundus photographs (CFPs) of a patient without DR (left) and a patient with signs of DR (right). In the CFP of the patient with signs of DR (right), one example each of hemorrhage, exudate, and a microaneurysm are highlighted. Both examples have been selected from the Kaggle DR dataset.47 b Schematic of the Diabetic Retinopathy Severity Scale (DRSS) established by the Early Treatment Diabetic Retinopathy Study (ETDRS) group to measure DR worsening over time. c Schematic of the two-phase modeling to detect two-step or more DRSS worsening over time. In phase I, field-specific Inception-v3 deep convolutional neural networks (DCNNs) called “field-specific DCNNs” or “pillars” are trained by means of transfer learning to predict whether the patient will progress two ETDRS DRSS steps. In phase II, the probabilities independently generated by the field-specific DCNNs are aggregated by means of random forestCC BY · PubMed Central
Fig. 2 Summary of the results for the prediction of two-step or more diabetic retinopathy progression at months 6, 12, and 24 using 7-field color fundus photographs of patients at baseline. AUC area under the curve, CI confidence interval, CV cross-validation, ROC receiver operating characteristic, SD standard deviation, SENS sensitivity, SPEC specificity
Fig. 2 Summary of the results for the prediction of two-step or more diabetic retinopathy progression at months 6, 12, and 24 using 7-field color fundus photographs of patients at baseline. AUC area under the curve, CI confidence interval, CV cross-validation, ROC receiver operating characteristic, SD standard deviation, SENS sensitivity, SPEC specificityCC BY · PubMed Central
Fig. 3 SHAP plots summarizing the pointwise and average contribution of each deep convolutional neural network (DCNN) to the random forest aggregation. SHAP plots outlining the pointwise contribution of each DCNN. In this example, the SHAP analysis is related to the five folds used for the prediction of DR progression at month 24 is showed. The DCNNs are ordered in importance from top to bottom. The naming convention of the DCNNs highlights the field (‘f1,’ ‘f2,’ etc.) and repetition (‘rep00,’ ‘rep01,’ etc.)
Fig. 3 SHAP plots summarizing the pointwise and average contribution of each deep convolutional neural network (DCNN) to the random forest aggregation. SHAP plots outlining the pointwise contribution of each DCNN. In this example, the SHAP analysis is related to the five folds used for the prediction of DR progression at month 24 is showed. The DCNNs are ordered in importance from top to bottom. The naming convention of the DCNNs highlights the field (‘f1,’ ‘f2,’ etc.) and repetition (‘rep00,’ ‘rep01,’ etc.)CC BY · PubMed Central
Fig. 4 Example of attribution maps placed side by side to the original test color fundus image. In each set, the original image is on the left and the attribution map is on the right. The attribution of the deep convolutional neural networks focuses mainly on microaneurysms, hemorrhages, and exudates. a Two examples of attribution maps for the model predicting diabetic retinopathy (DR) progression at month 6. b Two examples of attribution maps for the artificial intelligence model predicting DR progression at month 12. c Two examples of attribution maps for the model predicting DR progression at month 24
Fig. 4 Example of attribution maps placed side by side to the original test color fundus image. In each set, the original image is on the left and the attribution map is on the right. The attribution of the deep convolutional neural networks focuses mainly on microaneurysms, hemorrhages, and exudates. a Two examples of attribution maps for the model predicting diabetic retinopathy (DR) progression at month 6. b Two examples of attribution maps for the artificial intelligence model predicting DR progression at month 12. c Two examples of attribution maps for the model predicting DR progression at month 24CC 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 42 references

Research works printed but not linked: 4

  1. Zeiler M.D., Fergus R. in Computer Vision – ECCV 2014. ECCV 2014. Lecture Notes in Computer Science, vol 8689 (eds Fleet, D. et al.) (Springer, Cham, 2014).
  2. Brown, D. M. et al. Long-term outcomes of ranibizumab therapy for diabetic macular edema: the 36-month results from two phase III trials: RISE and RIDE. Ophthalmology 120, 2013–2022 (2013). · DOI · Paper page
  3. Springenberg, J. T., Dosovitskiy, A., Brox, T. & Riedmiller, M. Striving for simplicity: the all convolutional net. https://arxiv.org/abs/1412.6806 (2015). · arXiv
  4. Simonyan, K., Vedaldi, A. & Zisserman, A. Deep inside convolutional networks: visualising image classification models and saliency maps. https://arxiv.org/abs/1312.6034 (2014). · arXiv

Cited by

See all 305

Metadata

Authors
Arcadu, F., Benmansour, F., Maunz, A., Willis, J., Haskova, Z., & Prunotto, M.
Affiliations
  • F. Hoffmann-La Roche AG (CH)
  • University of Geneva (CH)
Corresponding author
Prunotto, Marco
Venue
npj Digital Medicine
Published
2019-09-20
Volume
2
Issue
1
Pages
92-92
ISSN
2398-6352 (ISSN-L)
OpenAlex
W2974726644
Cited by
305
References
42
Citations by year
2026: 20 · 2025: 32 · 2024: 58 · 2023: 33 · 2022: 55 · 2021: 53 · 2020: 50 · 2019: 3
FWCI
26.39
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Digital Medicine (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
Genentech
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, disease progression prediction, deep learning, color fundus images, microvascular abnormalities, retinal imaging

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: Deep learning algorithm predicts diabetic retinopathy progression in individual patients. https://app.cochat.ai/papers/10.1038/s41746-019-0172-3
The paper
Arcadu, F., Benmansour, F., Maunz, A., Willis, J., Haskova, Z., & Prunotto, M. (2019). Deep learning algorithm predicts diabetic retinopathy progression in individual patients. npj Digital Medicine. https://doi.org/10.1038/s41746-019-0172-3

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.