Abstract
Purpose :
To evaluate the use of deep learning system (DLS) as the grading tool, as compared to the human graders, to evaluate the association between referable diabetic retinopathy (DR) and systemic vascular risk factors including age, gender, duration of diabetes, glycated hemoglobin A1c (HbA1c), blood pressure, body mass index, total cholesterol and triglycerides
Methods :
This is a cross-sectional cohort study involving 157,798 images (78,883 eyes, 14,929 patients) captured from 4 multi-ethnic community-based and population-based studies over a 12- year period, evaluating the diagnostic performance of DLS, a novel research grading tool for referable DR. The reference standard for community-based screening program and population-based studies were a retinal specialist and professional graders, respectively.
Results :
The relationship between diabetes-related risk factors and DLS-generated referable DR were comparable to those generated by human graders (p>0.05) across all multi-ethnic datasets. Three systemic risk factors, including duration of diabetes, HbA1c and systolic blood pressure, were found to be statistical significantly (p<0.01) associated with referable DR diagnosed by DLS and the human graders in most datasets. Divided into quartiles, increasing DR severity was found to be significantly associated with increasing age (p<0.001), diabetes duration (p<0.001), HbA1c (p<0.001) and systolic blood pressure (p<0.001) in all 4 datasets.
Conclusions :
This study highlighted the clinical applicability and high public health significance of using DLS as a robust alternative grading tool, potentially saving tremendous manpower, infrastructure, costs and grading turnover time in many large community- and population-based cohort studies in the future, especially for those developing countries with finite resources.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.