Mass AI screening for diabetic retinopathy in Ukraine during the war.

Abstract

PURPOSE

To evaluate accuracy of AI-driven mass screening of diabetic retinopathy in Ukraine in wartime.

SETTING / VENUE

Prospective, open-label, observational study was coducted by State Institution "The Filatov Institute of Eye Diseases and Tissue Therapy of the NAMS of Ukraine".

METHODS

This study of mass screening of DR using an AI-based software platform was conducted from August 2023 till February 2024 during Russia invasion of Ukraine. It was based at the sites of healthcare facilities in the Zakarpatska oblast. One thousand six hundred twelve diabetics (3224 eyes) were involved in the study. In addition to residents of the Zakarpatska oblast, internally displaced persons from the Eastern regions of Ukraine were included in the study. Color fundus images were obtained using a nonmydriatic fundus camera with a 45-degree field of view. All fundus images were analyzed using the artificial intelligence (AI)-based software platform Retina-AI CheckEye©. Additonally every photo was independently verified by two ophthalmologists to confirm detection of DR. Receiver operating characteristic (ROC) curve analysis was performed to determine the sensitivity and specificity of the DR diagnosis method.

RESULTS

Signs of DR in at least one eye were detected in 564 diabetics or 35% of the diabetics. No DR signs were detected in 645 individuals (40% of total study subjects). In 806 eyes (25% of total eyes), the results were not obtained due to the features of the optical media and presence of certain eye diseases (in most cases, unilateral cataract, too narrow pupil, corneal opacity). This trial found 93% sensitivity and 88% specificity for the Retina-AI CheckEye-assisted detection of DR. 169 diabetics (30% of persons with detected DR) learned for the first time that they had diabetic retinopathy.

CONCLUSIONS

An AI-based software platform, Retina-AI CheckEye©, has been for the first time developed in Ukraine. The platform was demonstrated to have a high accuracy (93% sensitivity and 88% specificity) in detecting DR. This system may be used during wartime for the mass screening of DR.