Assessment of OCT angiography image quality in clinical studies

Abstract

Purpose : Artifacts of OCT angiography (OCTA) have been extensively described, however, there have been no reports of the prevalence of critical artifacts that negatively affect the reliability of vessel density (VD) output. We performed a retrospective, observational study to assess the prevalence and severity of artifacts in OCTA images in clinical studies.

Methods : OCTA images from 88 eyes (44 patients) were randomly selected from patients with diabetic retinopathy analyzed at the Fundus Photograph Reading Center, University of Wisconsin. Both 3mm and 6mm images were assessed for each eye (n=135) when available. Images were independently graded by two graders for presence and severity of grid decentration, automated segmentation error, eye movement, blink, defocus, shadow, z-offset, tilt, and projection. Adjudication was performed by a third grader in cases of discrepancy. Critical artifacts were those that created unreliable VD outputs (Figure 1). The cause for artifact was categorized as technical, patient related, both, or unknown.

Results : Of the 135 images assessed, 62% were graded as unreliable for VD due to at least one critical artifact. Defocus (54%), shadow (48%), and eye movement (30%) were the most common critical artifacts. Half of all images with unreliable VD output (42/84) had more than one critical artifact present in the same image. Frequency of unreliable images was similar between 6mm and 3mm scans (64% vs 60%; p = 0.633). Of the 6mm scans, defocus (69%), shadow (58%), and tilt (27%) were the most common critical artifacts. Of the 3mm scans, eye movement (42%), defocus (33%), and shadow (33%) were the most prevalent critical artifacts. Artifacts resulting from technical errors were present in 73% of images, while patient attributed artifacts such as eye movement were present in 58% of images.

Conclusions : Critical artifacts affecting quantitative measurement of vessel density are highly prevalent in OCTA imaging of clinical studies. Improving technical parameters would greatly reduce the prevalence of unreliable images.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.