강화학습 기반 인슐린 주입량 추천 알고리즘

Abstract

We propose an insulin-dosing algorithm to manage diabetes patients’ blood glucose level and prevent hypoglycemia. To effectively personalize the insulin suggestions, we used reinforcement learning, a type of artificial intelligence, and validated the suggestions in-silico with ten virtual patients under realistic meal scenarios. The meal scenarios consist of three meal sizes and three meal times. The time-in-range of the algorithm was 85.56%, and the blood glucose level was measured using continuous glucose monitoring sensor. These results can help prevent hypoglycemia and improve the quality of life for diabetes patients.