Purchase this article with an account.
Sumana Sri Kommana, Nicole Mendez, Pooja Anand Padgaonkar, Lesley Wu, Bernard C Szirth, Albert S Khouri; Comparing the accuracy of automated retinal analysis software to manual reading for Diabetic Retinopathy detection in DM Type 1. Invest. Ophthalmol. Vis. Sci. 2016;57(12):5963. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To assess the accuracy of a novel automated retinal software vs. manual reading in assessing diabetic retinopathy in subjects with DM Type 1.
Color retinal fundus image encounters were analyzed using EyeArt™ (Eyenuk, Inc., Los Angeles, CA) automated DR screening software, which reports DR severity based on the ICDR (International Clinical Diabetic Retinopathy) scale and offers a score from 0-5, with higher scores signifying higher DR severity levels. EyeArt offers a final output of “refer” or “no refer” and is devised to provide a “refer” recommendation when it detects scores of moderate NPDR or higher on the ICDR scale. Ten images from a 20-year-old Hispanic male with type 1 diabetes for 16 years who was followed for 5 consecutive visits beginning in 2013 were included in the analysis. Color fundus images of both eyes were captured in non-mydriatic mode at an angle of 45° and a flash setting of 100 watt second using a Canon CR2 Plus AF retinal camera with an 18 Mp CMOS censor.A certified diabetic reader performed manual readings on all images. The images were analyzed for total number of lesions, which were further differentiated by type: dot, flame, and/or intraretinal microvascular abnormalities (IrMAs). A board certified ophthalmologist reviewed all images.
High-resolution images (5 OD, 5 OS) from 5 visits were analyzed by both EyeArt™ and manually by a certified diabetic reader. EyeArt recommended referral to a specialist for 9/10 images. The highest DR severity score was given to the images, which correspond to the visit with the most number of reported lesions. Software and manual analysis reports are provided in Table 1 and Table 2, respectively.
Our results show that EyeArt ™ software performed well when compared to manual assessment and has potential applications in screening for DR in young Type 1 DM. However, there are some limitations: for example, the retinal reflex found in young eyes can provide false positive counts, and the software does not output the various types of hemorrhages and their proximity to the macula and optic nerve head. Future versions of the software might include input of blood pressure as well as HbA1C levels to allow for greater sensitivity of the triage session. Planned studies will include analysis of a greater number of Type 1 DM subjects over a wider age and follow-up distribution.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
This PDF is available to Subscribers Only