Artificial Intelligence Allows Higher Detection of Diabetic Retinopathy in Youths

Autonomous artificial intelligence (AI) systems can effectively uncover diabetic retinopathy (DR) in patients between 5 years and 12 years old, according to findings published in Diabetes Care. 

Researchers conducted a prospective study wherein 310 pediatric patients received diabetic eye exams using a nonmydriatic fundus camera with an autonomous AI system to screen for DR. The study was carried out at a multidisciplinary pediatric diabetes clinic affiliated with the Johns Hopkins School of Medicine. 

After the vision screening using AI, retinal specialists, who were masked to AI output, compared their consensus with the sensitivity, specificity, and diagnosability obtained by the automated system. The researchers measured adherence to screening guidelines before and after AI was implemented. Of the 310 participants, 4.2% had DR, and the diagnosability of AI was 97.5% (n=302). Accuracy of the detection of more-than-mild DR was 85.7% for sensitivity and 79.3% for specificity.

The study explains that DR is a leading cause of vision loss in young adults throughout the world. However, despite the recommendation for screenings in children and adolescents, many never adhere to it. The study included regular patients of the clinic between December 2018 and November 2019 who had type 1 diabetes (T1D), type 2 diabetes (T2D), or cystic fibrosis-related diabetes.

The researchers explain that, depending on the type of diabetes diagnosed, the American Diabetes Association (ADA) recommends screening patients for DR either immediately or within 3 years. 

Studies show that 35% to 72% of youth with diabetes undergo recommended ophthalmic exams in accordance with clinical practice guidelines, the study explains. “Furthermore, minority youth and children from lower socioeconomic backgrounds are less likely to undergo recommended screening compared with their White counterparts, even with insurance coverage.”

The researchers in this study saw that adherence to screening guidelines improved with the use of AI, and they concluded that use of a nonmydriatic fundus camera was safe and effective. They are hopeful that the implementation of AI, especially in multidisciplinary clinics such as the site of this study, will allow children and adolescents  with diabetes to be screened for DR with more ease. 

“The performance in this pediatric study was compared with clinician grading and shows noninferiority to the FDA’s thresholds,” the study says. “There is also the advantage of an immediate result associated with AI in addition to time and cost savings to the patient.”

The primary limitation of the study was the low prevalence of DR in the participants, which could have limited conclusions on specificity and sensitivity.

Reference

Wolf R, Liu T, Alvin T, et al. The SEE study: safety, efficacy, and equity of implementing autonomous artificial intelligence for diagnosing diabetic retinopathy in youth. Diabetes Care. doi:10.2337/dc20-1671.