Retinal AI System Can Detect SBI, Predict Future Stroke Risk

Background: Silent brain infarctions (SBIs) are common subclinical manifestations of cerebrovascular disease and are associated with increased risk of future symptomatic stroke. Detection currently relies on brain imaging, which is impractical for large-scale screening. Objective: To evaluate DeepRETStroke, a retinal image–based deep learning system designed to detect SBI and predict stroke risk using fundus photography. Design: Development and validation study. Methods: Investigators validated data across international cohorts comprising 213,762 retinal images from China, Singapore, Malaysia, the United States, the United Kingdom, and Denmark. They pretrained the model using 895,640 retinal photographs. The algorithm was designed to perform 3 principal tasks: detection of SBI, prediction of incident stroke, and prediction of recurrent stroke. Results: For SBI detection, the fundus model achieved an area under the curve (AUC) of 0.797, outperforming models based solely on conve more...