Deep learning for gradability classification of handheld, non-mydriatic retinal images

Paul Nderitu; Joan Nunez do Rio; Rajiv Rasheed; Rajiv Raman; Ramachandran Rajalakshmi; Christos Bergeles; Sobha Sivaprasad

doi:10.1038/s41598-021-89027-4

Deep learning for gradability classification of handheld, non-mydriatic retinal images

Paul Nderitu, Joan Nunez do Rio, Rajiv Rasheed, Rajiv Raman, Ramachandran Rajalakshmi, Christos Bergeles, Sobha Sivaprasad

Vision Research Foundation India

Research output: Contribution to journal › Article › peer-review

9 Citations (Scopus)

Abstract

Screening effectively identifies patients at risk of sight-threatening diabetic retinopathy (STDR) when retinal images are captured through dilated pupils. Pharmacological mydriasis is not logistically feasible in non-clinical, community DR screening, where acquiring gradable retinal images using handheld devices exhibits high technical failure rates, reducing STDR detection. Deep learning (DL) based gradability predictions at acquisition could prompt device operators to recapture insufficient quality images, increasing gradable image proportions and consequently STDR detection. Non-mydriatic retinal images were captured as part of SMART India, a cross-sectional, multi-site, community-based, house-to-house DR screening study between August 2018 and December 2019 using the Zeiss Visuscout 100 handheld camera. From 18,277 patient eyes (40,126 images), 16,170 patient eyes (35,319 images) were eligible and 3261 retinal images (1490 patient eyes) were sampled then labelled by two ophthalmologists. Compact DL model area under the receiver operator characteristic curve was 0.93 (0.01) following five-fold cross-validation. Compact DL model agreement (Kappa) were 0.58, 0.69 and 0.69 for high specificity, balanced sensitivity/specificity and high sensitivity operating points compared to an inter-grader agreement of 0.59. Compact DL gradability model performance was favourable compared to ophthalmologists. Compact DL models can effectively classify non-mydriatic, handheld retinal image gradability with potential applications within community-based DR screening.

Original language	English
Article number	9469
Journal	Scientific Reports
Volume	11
Issue number	1
DOIs	https://doi.org/10.1038/s41598-021-89027-4
Publication status	Published - 4 May 2021

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1038/s41598-021-89027-4

Cite this

@article{3a0afcb2ec3041bb861ce8e4245c2133,

title = "Deep learning for gradability classification of handheld, non-mydriatic retinal images",

abstract = "Screening effectively identifies patients at risk of sight-threatening diabetic retinopathy (STDR) when retinal images are captured through dilated pupils. Pharmacological mydriasis is not logistically feasible in non-clinical, community DR screening, where acquiring gradable retinal images using handheld devices exhibits high technical failure rates, reducing STDR detection. Deep learning (DL) based gradability predictions at acquisition could prompt device operators to recapture insufficient quality images, increasing gradable image proportions and consequently STDR detection. Non-mydriatic retinal images were captured as part of SMART India, a cross-sectional, multi-site, community-based, house-to-house DR screening study between August 2018 and December 2019 using the Zeiss Visuscout 100 handheld camera. From 18,277 patient eyes (40,126 images), 16,170 patient eyes (35,319 images) were eligible and 3261 retinal images (1490 patient eyes) were sampled then labelled by two ophthalmologists. Compact DL model area under the receiver operator characteristic curve was 0.93 (0.01) following five-fold cross-validation. Compact DL model agreement (Kappa) were 0.58, 0.69 and 0.69 for high specificity, balanced sensitivity/specificity and high sensitivity operating points compared to an inter-grader agreement of 0.59. Compact DL gradability model performance was favourable compared to ophthalmologists. Compact DL models can effectively classify non-mydriatic, handheld retinal image gradability with potential applications within community-based DR screening.",

author = "Paul Nderitu and {Nunez do Rio}, Joan and Rajiv Rasheed and Rajiv Raman and Ramachandran Rajalakshmi and Christos Bergeles and Sobha Sivaprasad",

note = "Funding Information: The study is approved by the Indian Council of Medical Research (ICMR)/Health Ministry Screening Committee (HMSC). The study was conducted in accordance with the tenets of the Declaration of Helsinki. Each patient provided informed consent for participation in the study. The ORNATE India project is a 4-year Global Challenge Research Fund (GCRF) and UK Research and Innovation (UKRI) funded multicentre study whose ambition is to build research capacity and capability to tackle DR related visual impairment in India and the UK31. One key aim is to initiate community-based DR screening in India using a low-cost, non-mydriatic portable camera (SMART India study) and train DL models to assist in the automated detection of DR31. One of the first steps to achieving this goal is the development of an image quality assessment tool that can assist device operators in the acquisition of gradable retinal images31. Funding Information: This study was funded in part by the UK Research and Innovation (UKRI): Global Challenge Research Fund (GCRF) [MR/P027881/1]. The funder did not influence the conduct of this study including data collection, management, analysis or interpretation. Manuscript preparation and review was independent to the funder. Publisher Copyright: {\textcopyright} 2021, The Author(s). Copyright: Copyright 2021 Elsevier B.V., All rights reserved.",

year = "2021",

month = may,

day = "4",

doi = "10.1038/s41598-021-89027-4",

language = "English",

volume = "11",

journal = "Scientific Reports",

issn = "2045-2322",

publisher = "Nature Publishing Group",

number = "1",

}

TY - JOUR

T1 - Deep learning for gradability classification of handheld, non-mydriatic retinal images

AU - Nderitu, Paul

AU - Nunez do Rio, Joan

AU - Rasheed, Rajiv

AU - Raman, Rajiv

AU - Rajalakshmi, Ramachandran

AU - Bergeles, Christos

AU - Sivaprasad, Sobha

N1 - Funding Information: The study is approved by the Indian Council of Medical Research (ICMR)/Health Ministry Screening Committee (HMSC). The study was conducted in accordance with the tenets of the Declaration of Helsinki. Each patient provided informed consent for participation in the study. The ORNATE India project is a 4-year Global Challenge Research Fund (GCRF) and UK Research and Innovation (UKRI) funded multicentre study whose ambition is to build research capacity and capability to tackle DR related visual impairment in India and the UK31. One key aim is to initiate community-based DR screening in India using a low-cost, non-mydriatic portable camera (SMART India study) and train DL models to assist in the automated detection of DR31. One of the first steps to achieving this goal is the development of an image quality assessment tool that can assist device operators in the acquisition of gradable retinal images31. Funding Information: This study was funded in part by the UK Research and Innovation (UKRI): Global Challenge Research Fund (GCRF) [MR/P027881/1]. The funder did not influence the conduct of this study including data collection, management, analysis or interpretation. Manuscript preparation and review was independent to the funder. Publisher Copyright: © 2021, The Author(s). Copyright: Copyright 2021 Elsevier B.V., All rights reserved.

PY - 2021/5/4

Y1 - 2021/5/4

N2 - Screening effectively identifies patients at risk of sight-threatening diabetic retinopathy (STDR) when retinal images are captured through dilated pupils. Pharmacological mydriasis is not logistically feasible in non-clinical, community DR screening, where acquiring gradable retinal images using handheld devices exhibits high technical failure rates, reducing STDR detection. Deep learning (DL) based gradability predictions at acquisition could prompt device operators to recapture insufficient quality images, increasing gradable image proportions and consequently STDR detection. Non-mydriatic retinal images were captured as part of SMART India, a cross-sectional, multi-site, community-based, house-to-house DR screening study between August 2018 and December 2019 using the Zeiss Visuscout 100 handheld camera. From 18,277 patient eyes (40,126 images), 16,170 patient eyes (35,319 images) were eligible and 3261 retinal images (1490 patient eyes) were sampled then labelled by two ophthalmologists. Compact DL model area under the receiver operator characteristic curve was 0.93 (0.01) following five-fold cross-validation. Compact DL model agreement (Kappa) were 0.58, 0.69 and 0.69 for high specificity, balanced sensitivity/specificity and high sensitivity operating points compared to an inter-grader agreement of 0.59. Compact DL gradability model performance was favourable compared to ophthalmologists. Compact DL models can effectively classify non-mydriatic, handheld retinal image gradability with potential applications within community-based DR screening.

AB - Screening effectively identifies patients at risk of sight-threatening diabetic retinopathy (STDR) when retinal images are captured through dilated pupils. Pharmacological mydriasis is not logistically feasible in non-clinical, community DR screening, where acquiring gradable retinal images using handheld devices exhibits high technical failure rates, reducing STDR detection. Deep learning (DL) based gradability predictions at acquisition could prompt device operators to recapture insufficient quality images, increasing gradable image proportions and consequently STDR detection. Non-mydriatic retinal images were captured as part of SMART India, a cross-sectional, multi-site, community-based, house-to-house DR screening study between August 2018 and December 2019 using the Zeiss Visuscout 100 handheld camera. From 18,277 patient eyes (40,126 images), 16,170 patient eyes (35,319 images) were eligible and 3261 retinal images (1490 patient eyes) were sampled then labelled by two ophthalmologists. Compact DL model area under the receiver operator characteristic curve was 0.93 (0.01) following five-fold cross-validation. Compact DL model agreement (Kappa) were 0.58, 0.69 and 0.69 for high specificity, balanced sensitivity/specificity and high sensitivity operating points compared to an inter-grader agreement of 0.59. Compact DL gradability model performance was favourable compared to ophthalmologists. Compact DL models can effectively classify non-mydriatic, handheld retinal image gradability with potential applications within community-based DR screening.

UR - http://www.scopus.com/inward/record.url?scp=85105373777&partnerID=8YFLogxK

U2 - 10.1038/s41598-021-89027-4

DO - 10.1038/s41598-021-89027-4

M3 - Article

SN - 2045-2322

VL - 11

JO - Scientific Reports

JF - Scientific Reports

IS - 1

M1 - 9469

ER -

Deep learning for gradability classification of handheld, non-mydriatic retinal images

Abstract

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this