Automatic identification of referral-warranted diabetic retinopathy using deep learning on mobile phone images

Abstract

Purpose : To evaluate the performance of a previously published and validated artificial intelligence algorithm for detecting diabetic retinopathy (DR) on low-resolution fundus images acquired with a mobile phone and indirect ophthalmoscope lens adapter

Methods : A validated fully automated DR deep learning algorithm trained on 75,137 fundus images was tested on a dataset of 109 usable fundus images acquired from two previously published studies. Images were extracted from live video screenshots from fundus examinations using the EyeGo device and exported as a screenshot from live video clips filmed at 1080p resolution. Each image was graded twice by a board-certified ophthalmologist and compared to the output of the algorithm, which classified each image as having referral-warranted DR (moderate NPDR or worse) or no referral-warranted DR.

Results : Our algorithm achieved a 0.92 Area Under the Curve (AUC) with an 88% sensitivity and 94% specificity for detecting referral-warranted DR on mobile phone acquired fundus photos using a commercially available lens adapter.

Conclusions : The ability to screen patients with low-cost mobile devices has the potential to exponentially extend the reach of eye care and can serve the world's diabetic population. We showed that a fully data-driven artificial intelligence-based grading algorithm can be used to screen
fundus photos taken from mobile devices and identify with high reliability which cases should be referred to an
ophthalmologist for further evaluation and treatment. The implementation of such an algorithm on a global basis could
drastically reduce the rate of vision-loss attributed to DR.

This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.