Purchase this article with an account.
Kerry Cotter, Kristina Holbrook, Paul Andrew Yates; Incidental findings identified through diabetic retinopathy screening and potential impact on computer automated diabetic retinopathy reading algorithms. Invest. Ophthalmol. Vis. Sci. 2014;55(13):4824.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Tele-ophthalmic screening is a widely accepted tool for detecting diabetic retinopathy (DR). However, additional ocular pathology is often identified on screening images. In this study, we sought to evaluate the incidence of other retinal conditions identified on DR screening images. These findings may impact the ability of computer automated DR reading algorithms to substitute for human readers and accurately detect all patients at risk from vision loss due to ophthalmic disease.
Over a 24-month period, adult patients (N = 1267) with diabetes mellitus who had not received a dilated eye exam within the prior 12 months, underwent tele-ophthalmic screening for DR at University Medical Associates (UMA), a primary care practice within the University of Virginia Health System. A nurse obtained a single 45° macula-centered image of each eye using a refurbished non-mydriatic (Topcon TRC-45N) camera with a Canon T2i DSLR-back. The images were read by an ophthalmologist and graded for presence of DR. Incidental pathology was also recorded.
Diabetic retinopathy was detected in at least one eye in 255 (20%) subjects, an increased cup-to-disc ratio in 202 (16%) subjects, drusen or pigmentary changes associated with age-related macular degeneration (AMD) in 101 (8%) subjects, and other pathologies in 105 (8%) subjects. The most common “other” ocular pathologies identified were choroidal nevus, vascular tortuosity, and epiretinal membrane. Figure 1 shows a suboptimal screening image depicting an increased cup-to-disc ratio in a patient determined on follow-up clinical exam to have glaucoma. This indicates even poor quality images may demonstrate important retinal pathology.
Tele-ophthalmic screening can be used to identify co-morbid diseases other than DR. In particular, ocular findings consistent with glaucoma and AMD occurred in this diabetic population in total, at a higher frequency than DR itself. Even photos deemed “unreadable” for purposes of detecting DR, may capture important pathology. It is important automated screening algorithms identify these findings to avoid missing treatable eye disease, particularly in known at-risk populations. Further evaluation is needed to determine the sensitivity of disease detection via tele-ophthalmic screening in this population compared to a standard dilated fundus exam.
This PDF is available to Subscribers Only