Sensitivity and Specificity of Smartphone-Based Retinal Imaging for Diabetic Retinopathy: A Comparative Study

Abstract

Purpose: To determine the sensitivity and specificity of a smartphone-based fundus camera, the Remidio Fundus on Phone (FOP; Remidio Innovative Solutions Pvt. Ltd., Bengaluru, India) in detecting diabetic retinopathy (DR) compared with a conventional tabletop fundus camera and clinical examination.

Design: Cross-sectional, single-site, instrument validation study.

Participants: Consecutive patients with diabetes who had no DR (n = 55 eyes), mild to moderate nonproliferative diabetic retinopathy (NPDR; n = 70 eyes), severe NPDR (n = 46 eyes), proliferative diabetic retinopathy (PDR; n = 62 eyes), and diabetic macular edema (DME; n = 44 eyes).

Methods: All participants underwent a dilated examination to determine the grade of DR. Then all participants had mydriatic 45° fundus photographs obtained from three fields of view with the Remidio FOP and a Topcon tabletop fundus camera (Topcon Medical Systems, Inc., Oakland, NJ). Two masked retina specialists graded images for DR and DME, using National Health Service guidelines as well as for image quality using predefined criteria.

Main outcome measure: Sensitivity and specificity of the Remidio FOP for the detection of any DR compared to clinical examination.

Results: One hundred thirty-five subjects (233 eyes) were recruited for the study. Compared with the reference clinical examination, using images from the Remidio FOP, graders 1 and 2 reported a sensitivity of 93.1% (95% confidence interval [CI] = 88.3-96.4) and 94.3% (95% CI = 89.7-97.2) and a specificity of 89.1% (95% CI = 68.2-92.2) and 94.5% (95% CI = 84.9-98.9), respectively, in identifying any DR (κ = 0.55; 95% CI = 0.50-0.57). With images from the Topcon camera, graders reported similar sensitivities and specificities with marginally better agreement (κ = 0.68; 95% CI = 0.67-0.73). The sensitivity of detecting DR gradually increased from R1 to R3 disease using both cameras. Both graders classified significantly fewer images as "ungradable" (2.6%-4.3% for Topcon vs. 1.7%-2.1% for Remidio FOP) and more images as excellent from the Remidio FOP (59%-74%) than the Topcon device (52%-61%).

Conclusions: The Remidio FOP device was found to have high sensitivity and specificity for the detection of any grade of DR with good agreement between graders. The rate of ungradable images was acceptably low and image quality was marginally better with the Remidio FOP.