计算机科学
稳健性(进化)
血糖性
机器学习
背景(考古学)
调度(生产过程)
人工智能
医学
胰岛素
工程类
运营管理
内科学
古生物学
生物化学
化学
生物
基因
作者
Chiara Toffanin,Lalo Magni
标识
DOI:10.1016/j.conengprac.2023.105578
摘要
In the last decades, the Run-to-Run (R2R) approach revealed to be a very promising strategy to automatically adapt the insulin therapy in unsupervised context. One of the key parameters to apply these algorithms in a supervised context is the updating frequency: low frequency causes a poor glycemic control, while very frequent therapy adaptation is a time-consuming task for physicians and patients. In this work, three different run durations for the R2R algorithm are tested on the 100 adult patients of the UVA/Padova simulator: a 1-day, a 7-day and a 28-day. The 1-day and 7-day R2Rs show comparable performances, while the 28-day R2R presents some suboptimal behavior. Starting from this preliminary analysis, a new clinical protocol for the insulin therapy update, including variable update frequencies personalized for each patient, is designed in order to be safe for the patient and clinically feasible for the medical staff. The proposed protocol shows promising results on a 6-month scenario tested in silico in robustness conditions.
科研通智能强力驱动
Strongly Powered by AbleSci AI