Prediction of systemic biomarkers from retinal photographs: development and validation of deep-learning algorithms

Tyler Hyungtaek Rim,Geunyoung Lee,Youngnam Kim,Yih-Chung Tham,Chan Joo Lee,Su Jung Baik,Young Ah Kim,Marco Yu,Mihir Deshmukh,Byoung Kwon Lee,Sungha Park,Hyeon Chang Kim,Charumathi Sabayanagam,Daniel S W Ting,Ya Xing Wang,Jost B Jonas,Sung Soo Kim,Tien Yin Wong,Ching-Yu Cheng

doi:10.1016/s2589-7500(20)30216-8

Abstract

The application of deep learning to retinal photographs has yielded promising results in predicting age, sex, blood pressure, and haematological parameters. However, the broader applicability of retinal photograph-based deep learning for predicting other systemic biomarkers and the generalisability of this approach to various populations remains unexplored. With use of 236 257 retinal photographs from seven diverse Asian and European cohorts (two health screening centres in South Korea, the Beijing Eye Study, three cohorts in the Singapore Epidemiology of Eye Diseases study, and the UK Biobank), we evaluated the capacities of 47 deep-learning algorithms to predict 47 systemic biomarkers as outcome variables, including demographic factors (age and sex); body composition measurements; blood pressure; haematological parameters; lipid profiles; biochemical measures; biomarkers related to liver function, thyroid function, kidney function, and inflammation; and diabetes. The standard neural network architecture of VGG16 was adopted for model development. In addition to previously reported systemic biomarkers, we showed quantification of body composition indices (muscle mass, height, and bodyweight) and creatinine from retinal photographs. Body muscle mass could be predicted with an R2 of 0·52 (95% CI 0·51-0·53) in the internal test set, and of 0·33 (0·30-0·35) in one external test set with muscle mass measurement available. The R2 value for the prediction of height was 0·42 (0·40-0·43), of bodyweight was 0·36 (0·34-0·37), and of creatinine was 0·38 (0·37-0·40) in the internal test set. However, the performances were poorer in external test sets (with the lowest performance in the European cohort), with R2 values ranging between 0·08 and 0·28 for height, 0·04 and 0·19 for bodyweight, and 0·01 and 0·26 for creatinine. Of the 47 systemic biomarkers, 37 could not be predicted well from retinal photographs via deep learning (R2≤0·14 across all external test sets). Our work provides new insights into the potential use of retinal photographs to predict systemic biomarkers, including body composition indices and serum creatinine, using deep learning in populations with a similar ethnic background. Further evaluations are warranted to validate these findings and evaluate the clinical utility of these algorithms. Agency for Science, Technology, and Research and National Medical Research Council, Singapore; Korea Institute for Advancement of Technology.

Highlights

The retina is the only organ that allows direct, noninvasive, in-vivo visualisation of the microvasculature and neural tissues
In addition to previously reported systemic biomarkers, we showed quantification of body composition indices and creatinine from retinal photographs
Using UK Biobank data, Poplin and colleagues showed that deeplearning models could predict six cardiovascular risk factors using only retinal photographs with reasonable accuracy.[5]

Summary

Introduction

The retina is the only organ that allows direct, noninvasive, in-vivo visualisation of the microvasculature and neural tissues It affords a unique opportunity for the non-invasive detection of systemic vascular and neurological diseases.[1] In recent decades, our under standing of retina–systemic disease relationships has relied on classic epidemiological studies based on obser vable, human-defined retinal features (eg, retinop athy or retinal vascular calibre).[2] The potential discovery of unobservable retinal features associated with systemic diseases has been enhanced by advances in artificial intellig ence technology, deep learning.[3] Deep learning can be used to predict many systemic biomarkers using retinal photographs, obviating the need for observable, precharacterised retinal features.[4] Using UK Biobank data, Poplin and colleagues showed that deeplearning models could predict six cardiovascular risk factors using only retinal photographs with reasonable accuracy.[5] Using these photographs, deep learning could accurately predict features such as age and sex, which otherwise cannot be identified by human eyes alone. Several studies on deep learning-predicted systemic conditions from retinal photographs have been published.[7,8,9,10,11]

Methods

Results

Conclusion