AI tool has potential to expand diabetic retinopathy screening programs
Theia, an AI-based diabetic retinopathy screening tool developed by Toku Eyes, could augment New Zealand’s diabetic retinopathy screening program in grading efficacy and increased patient representation, according to a study.
“AI-based algorithms that can reliably detect DR [diabetic retinopathy] in retinal images and provide instantaneous reporting with high diagnostic accuracy could significantly improve the earlier detection of DR,” Toku Eyes co-founder and CEO Ehsan VaghefiPhD, and colleagues wrote in Eye. “By enabling specialist-level diagnostics to be provided to multiple peripheral sites simultaneously, these algorithms also have the potential to significantly increase access to and lower the cost of screening for DR.”
In a prospective, multicenter trial, Vaghefi and colleagues used Theia to analyze images from 900 adult participants in the New Zealand Diabetic Screening Program, of whom 243 had a referable disease and 657 had no or minimal disease. Fundus images were gathered from two separate clinics — a large urban tertiary clinic and a rural optometric practice — between January and April of 2021. According to the study, a variety of cameras and locations were used to be more representative of New Zealand’s DR population.
Images were de-identified and shared with three DR specialists (and in some instances, one additional retinal specialist) to be graded and compared with Theia. The grading system was based on the New Zealand grading system at the patient level, which has both nonreferable and referable classification, as well as healthy, mild, more than mild and sight-threatening categories.
Researchers used kappa statistic and percentage agreement to assess agreement between specialists and Theia. The primary outcome was sensitivity and specificity of Theia in detecting retinopathy.
Vaghefi and colleagues found that Theia had an overall accuracy of 98%, with 100% sensitivity and 98.18% specificity in nonreferable vs. referable grading. The level of agreement between the AI tool and the specialists was also very high, according to the study.
In the analysis of disease severity, Theia tended to over grade the degree of retinopathy, but did not miss any instances of disease. The level of agreement between the specialists and Theia ranged from k values of 0.96 to 0.75.
“This multicenter prospective trial demonstrates that Theia is capable of detecting DR ... with a very high degree of accuracy, while providing a high level of granularity in grading,” Vaghefi and colleagues wrote. “With appropriate oversight and audit, these results indicate that Theia could be safely deployed within established diabetic screening programs to augment the expertise of the clinicians, increasing overall screening capacity while reducing costs per unit screen.”
Perspective
Back to Top
This study showcases software called Theia who, in Greek mythology, is the Titan goddess of sight. I’ll write this perspective focused on (ophthalmic-related) artificial intelligence software in general, rather than on ones with such lofty eponyms.
I’ve been a (human) grader of diabetic retinopathy in retinal photographs for almost 20 years as a part of my hospital’s tele-eye care program. I think it’s folly to think that we eye care providers will forever outperform a well-programed AI-based model to screen for diabetic retinopathy. This study is an important part of the process of validating such software for clinical use.
As practicing clinicians, we should not only be informed of this but start using these opportunities to improve the way we care for patients. This specific technology is essentially used as a point-of-care opportunity; the patient is informed in near-real time if the screening results are concerning. The questions the clinician should ask then are: To whom is the next call made? And how do I get on the other end of that call? This study used graders in “urban tertiary clinics” but also in “provincial optometric practices.” There is room for both patients and clinicians to benefit as the onward march of technology continues.
Collapse
Read more about