Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2022 Jul 1;12(1):11196.
doi: 10.1038/s41598-022-15491-1.

Automated image curation in diabetic retinopathy screening using deep learning

Paul Nderitu  1   2 , Joan M Nunez do Rio  3 , Ms Laura Webster  4 , Samantha S Mann  4   5 , David Hopkins  6   7 , M Jorge Cardoso  8 , Marc Modat  8 , Christos Bergeles  8 , Timothy L Jackson  3   9
Affiliations

Automated image curation in diabetic retinopathy screening using deep learning

Paul Nderitu et al. Sci Rep. .

Abstract

Diabetic retinopathy (DR) screening images are heterogeneous and contain undesirable non-retinal, incorrect field and ungradable samples which require curation, a laborious task to perform manually. We developed and validated single and multi-output laterality, retinal presence, retinal field and gradability classification deep learning (DL) models for automated curation. The internal dataset comprised of 7743 images from DR screening (UK) with 1479 external test images (Portugal and Paraguay). Internal vs external multi-output laterality AUROC were right (0.994 vs 0.905), left (0.994 vs 0.911) and unidentifiable (0.996 vs 0.680). Retinal presence AUROC were (1.000 vs 1.000). Retinal field AUROC were macula (0.994 vs 0.955), nasal (0.995 vs 0.962) and other retinal field (0.997 vs 0.944). Gradability AUROC were (0.985 vs 0.918). DL effectively detects laterality, retinal presence, retinal field and gradability of DR screening images with generalisation between centres and populations. DL models could be used for automated image curation within DR screening.

PubMed Disclaimer

Conflict of interest statement

P Nderitu, JM Nunez do Rio, L Webster, SS Mann, D Hopkins, MJ Cardoso, M Modat, C Bergeles has no conflicts of interest to declare. T Jacksons’ employer (King’s College Hospital) receives funding for participants enrolled on commercial clinical trials of diabetic retinopathy including THR149-002 (sponsor: OXURION), NEON NPDR (sponsor: BAYER), RHONE-X (sponsor: ROCHE) and ALTIMETER (sponsor: ROCHE). He has been paid for an expert clinical opinion by Kirkland and Ellis Solicitors, acting for REGENERON.

Figures

Figure 1
Figure 1
Automated image curation criteria. Automated image curation requires the detection of (1) laterality, (2) retinal presence (retinal vs non-retinal images), (3) retinal field (macula vs nasal vs other retinal fields) and (4) gradability which allows for the selection of gradable, 2-field retinal images of identifiable laterality for manual or automated DR grading.
Figure 2
Figure 2
Study dataset flow chart.
Figure 3
Figure 3
Single-output model receiver operating characteristic curves. 1Test set size = 1541 images, 2Test set size = 1479 images, 3Test set size = 1466 images, 4Test set size = 1427 images. AUROC: area-under-the receiver operating characteristic curve, ORF: other retinal field.
Figure 4
Figure 4
Multi-output model receiver operating characteristic curves. 1Test set size = 1541 images, 2Test set size = 1479 images, 3Test set size = 1466 images, 4Test set size = 1427 images. AUROC: area-under-the receiver operating characteristic curve, ORF: other retinal field.
Figure 5
Figure 5
Internal test single-output model pixel attribution maps. Integrated gradients pixel attributions: all models highlight the optic cup/disc within retinal images, especially model c. Models a, b and d also highlight the retinal vessels to varying degrees. Model b (non-retinal image) highlights the caruncle, lower tear meniscus, iris striations, conjunctival vessels, and corneal reflection. Model attributions relative to the true positive class in each image.
Figure 6
Figure 6
Proposed curation workflow. (a) Images get predictions for laterality and retinal presence (values indicate model predictions between 0 and 1) allowing for the exclusion of non-retinal images (e.g., anterior segment). (b) Images obtain retinal field and gradability predictions which allows for the exclusion of other retinal field images and for the selection of gradable images from macula or nasal fields by selecting the image with the highest gradable score (underlined). (c) The ‘best’ macula and nasal field with an identifiable laterality are then selected; these gradable, 2-field images are then suitable for subsequent manual or automated diabetic retinopathy grading. R: Retinal presence, OS: Left eye, N: Nasal, M: Macula, ORF: Other retinal field, G: Gradability.
See this image and copyright information in PMC

References

    1. Saeedi P, et al. Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: Results from the International Diabetes Federation Diabetes Atlas, 9(th) edition. Diabetes Res. Clin. Pract. 2019;157:107843. doi: 10.1016/j.diabres.2019.107843. - DOI - PubMed
    1. IDF. IDF Diabetes Atlas: Ninth Edition. (2019).
    1. Lee R, Wong TY, Sabanayagam C. Epidemiology of diabetic retinopathy, diabetic macular edema and related vision loss. Eye Vis. 2015;2:17. doi: 10.1186/s40662-015-0026-2. - DOI - PMC - PubMed
    1. Ting DS, Cheung GC, Wong TY. Diabetic retinopathy: Global prevalence, major risk factors, screening practices and public health challenges: a review. Clin. Exp. Ophthalmol. 2015;44:260–277. doi: 10.1111/ceo.12696. - DOI - PubMed
    1. Blindness, G. B. D. Vision Impairment, C. Vision Loss Expert Group of the Global Burden of Disease, S Causes of blindness and vision impairment in 2020 and trends over 30 years, and prevalence of avoidable blindness in relation to VISION 2020: The Right to Sight: an analysis for the Global Burden of Disease Study. Lancet Glob. Health. 2021;9:e144–e160. doi: 10.1016/S2214-109X(20)30489-7. - DOI - PMC - PubMed

Publication types