Abstract
Purpose :
SELENA+ (Singapore Eye Lesion Analyser Plus) is a deep learning artificial intelligence (AI) model that was trained on 76,370 retinal fundus images from 13,099 adults with diabetes. In this study, we evaluated the performance of SELENA+ for the detection of referable diabetic retinopathy (DR) in children and young people with diabetes (CYP).
Methods :
We retrospectively included CYP who attended DR screening annually at our institution between 2014 and 2023, during which their retinal fundus photos were taken and graded by trained professionals. Referable DR was defined as moderate non-proliferative DR or worse. Each of the retinal fundus image was presented to SELENA+ to evaluate its sensitivity and specificity for detecting referable DR in the CYP, and the area under the receiver operating characteristic (ROC) curve (AUC) was derived.
Results :
Among the 467 CYP included, 49% are male. 65% of subjects had type 1 diabetes and 35% had type 2 diabetes, with a median age of 10.8y [IQR 7.3y – 13.4y] at diagnosis and first DR screening age at 14.1y [11.4y – 16.2y]). A total of 8126 retinal fundus photos from 3396 visits were used to evaluate the performance of SELENA+ in detecting referable DR in CYP. On visit instances, DR was found in 279 (8.2%) eyes, and referable DR in 31 (0.9%) eyes. The AUC of SELENA+ for referable DR was 0.874 (95% CI, 0.795-0.947), with a sensitivity of 80.65% (62.53%-92.55%) at the specificity of 82.08% (80.74% - 83.36%).
Conclusions :
Clinically acceptable performance of SELENA+ for the detection of referable DR was found in this cohort of CYP, even when the AI model was trained in the adult population. Our study shows the potential application and adoption of such AI technology in screening CYP for DR. We plan to fine-tune the AI model specifically for use in pediatric patients by training it using more CYP and synthetic retinal fundus images as well as incorporating relevant clinical features and data.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.