Comparisons of Handheld Retinal Imaging Devices with Ultrawide Field (UWF) and Early Treatment Diabetic Retinopathy Study (ETDRS) Photographs for Determining Diabetic Retinopathy (DR) Severity

Abstract

Purpose : To compare DR severity identified on handheld retinal imaging with UWF and ETDRS photographs.

Methods : Mydriatic images of 225 eyes in 116 diabetic patients were taken with the Aurora (AU) handheld retinal imager and compared with stereoscopic UWF images and standard 7-field ETDRS photos taken at the same visit. Monoscopic 5-field imaging protocol was used for handheld images (centered on macula, optic nerve, and steered superior, inferior, and temporal). Images were evaluated for DR severity by certified graders using the International Classification for DR. Exact and within 1-step agreement between the two imaging modalities were determined, simple and weighted kappa statistics (K/Kw) were calculated. Grading discrepancies were adjudicated to determine the lesions and the sources of error.

Results : Distribution of DR severity by AU/ETDRS photos/UWF images (%): no DR 41.3/33.3/36.0, mild nonproliferative DR (NPDR) 18.7/20.4/17.8, moderate 10.2/14.2/10.7, severe in 16.4/11.6/15.1, proliferative DR (PDR) 13.3/20.4/20.4. All eyes were gradable by all modalities. Agreement between ETDRS and AU images was exact 65.8%, within 1-step 93.8%, k= 0.55 kw=0.75. Agreement between ETDRS and UWF was exact 70.6%, within 1-step 98.7%, k= 0.62 kw=0.81. Agreement between UWF and AU was exact agreement 68.0%, within 1-step 92.9%, k= 0.58 kw=0.76. Table 1 shows the cross-tabulation of DR severity by UWF and AU. Discrepancy of ≥2-steps between UWF and AU was observed in 16 eyes [8(50%) not in AU field, 6(37.5%) poor AU image quality, 2(12.5%) AU grader error) and the lesion causing the discrepancy was intraretinal microvascular abnormalities 5(31.2%), hemorrhages or microaneurysms 4 (25.0%), vitreous/preretinal hemorrhage 4(25.0%, figure 1), new vessels 3(18.8%)]. Using a referral threshold of moderate NPDR, 3(6.5%) eyes with PDR will be missed by AU due to limited field, while for a threshold of severe NPDR, 10(21.7%) eyes with PDR would be missed by AU due to image quality and field discrepancies.

Conclusions : Mydriatic 5-field handheld images using AU had moderate to strong agreement with UWF and ETDRS imaging. However, depending on the referral threshold, substantial levels of vision-threatening disease may be missed, suggesting that lower thresholds of referral are warranted if handheld devices are to be used.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.