Advanced Hypoglycemia Prevention Capabilities in Fuzzy Logic Artificial Pancreas

描述（由申请人提供）：本研究的最终技术目标是开发一种人工胰腺（AP）控制器，通过有效预防低血糖发作来改善代谢控制并降低血糖波动。比例-积分-微分控制（PID）和模型预测控制（MPC）已被广泛认为是有前途的候选人的葡萄糖控制。然而，PID和MPC方法依赖于模型，并且由于许多生理限制，例如，不可测量的膳食量和频繁的胰岛素敏感性变化。无模型方法，如模糊逻辑（FL）控制提供了一个不同的和有前途的方向，改善血糖控制。FL的另一个好处是与传统控制器相比，它所需的计算需求较低，这导致更小的CPU，更低的功耗要求和更小的电池，所有这些都导致更可用的AP系统。本研究旨在改善我们2010-11年JDRF资助的临床试验中使用的控制器的血糖控制和低血糖预防能力。具体而言，本研究的第1部分将使用UVA模拟器评估我们目前FL控制器的四个具体潜在改进：1）采用反应和预测算法的低血糖暂停（LGS）功能; 2）新型控制器个性化和在线适应功能; 3）针对快速与缓慢胰岛素反应者定制的给药矩阵;以及4）当在临床环境中遇到CGM传感器异常时，提高了控制器的安全性。在可能的情况下，我们将使用JDRF资助试验的临床数据来进一步验证新的控制器功能。初步研究表明，所提出的改进措施是可行的。本研究的第2部分发生在该资助的第二年，重点是在各种参数下对这些功能进行临床评价，以确定FL控制器的下一个主要配置。第2部分的技术问题是，计算机控制器的改变是否可以转化为人类的实际改善。主要目的是避免低血糖。这些研究的成功完成可能会导致一个完全闭环的，商业上可用的人工胰腺的发展。由此产生的FL控制器软件产品将提供给商业糖尿病医疗器械制造商，作为AP控制算法的选择。

DESCRIPTION (provided by applicant): The ultimate technical objective of this research is the development of an Artificial Pancreas (AP) controller that improves metabolic control and decreases glycemic excursions by robustly preventing hypoglycemic episodes. Proportional-integral-derivative control (PID) and model predictive control (MPC) have been widely considered to be promising candidate for glucose control. However, PID and MPC methods are dependent on models, and a good model for T1DM is not easy to develop because of a number of physiological limitations, e.g., unmeasured meal size and frequent insulin sensitivity variations. Model-free approaches, such as fuzzy logic (FL) control offer a different and promising direction for improved glycemic control. A further benefit of FL is the low computational needs it requires compared to traditional controllers, which lead to smaller CPUs, lower power requirements and smaller batteries, all of which lead to a more usable AP system. This research seeks to improve the blood sugar control and hypoglycemia prevention capabilities of the controller used in our 2010-11 JDRF-funded clinical trial. Specifically, Part 1 of this research will use the UVA simulator to evaluate four specific potential improvements to our present FL controller: 1), a Low Glucose Suspend (LGS) feature employing reactive and predictive algorithms; 2) novel controller personalization and online adaptation features; 3), dosing matrices tailored for fast vs. slow insulin responders; and 4), improved controller safety when encountering CGM sensor anomalies in the clinical environment. Where possible, clinical data from our JDRF-funded trial will be used to further validate the new controller features. The feasibility of the proposed improvements has been demonstrated by initial research. Part 2 of this research, occurring during the second year of this grant, focuses on the clinical evaluation of those features under various parameters, to establish the next major configuration of the FL controller. The technical question for Part 2 is whether the alterations in the controller in silico can be translated into actual improvement in humans. The primary goal is the avoidance of hypoglycemia. Successful completion of these studies could lead to the development of a fully closed loop, commercially available artificial pancreas. The resulting FL controller software product will be offered to commercial diabetes medical device manufactures as a choice for an AP control algorithm.