No 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.

Irreversible vision loss is a worldwide threat. Developing a computer-aided diagnosis system to detect retinal fundus diseases is extremely useful and serviceable to ophthalmologists. Early detection, diagnosis, and correct treatment could save the eye’s vision. Nevertheless, an eye may be afflicted with several diseases if proper care is not taken. A single retinal fundus image might be linked to one or more diseases. Age-related macular degeneration, cataracts, diabetic retinopathy, Glaucoma, and uncorrected refractive errors are the leading causes of visual impairment. Our research team at the center of excellence lab has generated a new dataset called the Retinal Fundus Multi-Disease Image Dataset 2.0 (RFMiD2.0). This dataset includes around 860 retinal fundus images, annotated by three eye specialists, and is a multiclass, multilabel dataset. We gathered images from a research facility in Jalna and Nanded, where patients across Maharashtra come for preventative and therapeutic eye care. Our dataset would be the second publicly available dataset consisting of the most frequent diseases, along with some rarely identified diseases. This dataset is auxiliary to the previously published RFMiD dataset. This dataset would be significant for the research and development of artificial intelligence in ophthalmology.

Diabetic retinopathy (DR) at risk of vision loss (referable DR) needs to be identified by retinal screening and referred to an ophthalmologist. Existing automated algorithms have mostly been developed from images acquired with high cost mydriatic retinal cameras and cannot be applied in the settings used in most low- and middle-income countries. In this prospective multicentre study, we developed a deep learning system (DLS) that detects referable DR from retinal images acquired using handheld non-mydriatic fundus camera by non-technical field workers in 20 sites across India. Macula-centred and optic-disc-centred images from 16,247 eyes (9778 participants) were used to train and cross-validate the DLS and risk factor based logistic regression models. The DLS achieved an AUROC of 0.99 (1000 times bootstrapped 95% CI 0.98–0.99) using two-field retinal images, with 93.86 (91.34–96.08) sensitivity and 96.00 (94.68–98.09) specificity at the Youden’s index operational point. With single field inputs, the DLS reached AUROC of 0.98 (0.98–0.98) for the macula field and 0.96 (0.95–0.98) for the optic-disc field. Intergrader performance was 90.01 (88.95–91.01) sensitivity and 96.09 (95.72–96.42) specificity. The image based DLS outperformed all risk factor-based models. This DLS demonstrated a clinically acceptable performance for the identification of referable DR despite challenging image capture conditions.

No abstract