Biomedical Control

biomed logo

Developing control systems for medical applications poses significant challenges, some of which are unique to this field. Physiological systems involve a multitude of interacting subsystems and networks, with multiple feedforward and feedback loops. The dynamics vary both between different individuals, and within the same individual over time. Measurement and estimation of both inter- and intra-patient variability poses a major challenge in the development of efficient controllers for biological and medical systems. Furthermore, many of the important states of these systems cannot be measured, and at times cannot even be estimated. The quantification of clinical objectives is another challenge, as they do not easily translate into the mathematical performance measures common in control systems theory.

Type 1 diabetes (T1D) is a prime candidate for the development of a treatment strategy using biomedical control. In this disease, the destruction of pancreatic beta cells leaves the body incapable of producing insulin, a hormone that is required for the glucose homeostasis feedback loop. The current standard of treatment is for people with T1D to measure their blood glucose concentration several times per day and manually deliver corresponding doses of insulin.

Our aim is to automate treatment by artificially recreating the glucose control feedback loop using a combination of medical devices. Specifically, novel approaches to patient characterization have been developed to predict glucose profiles under various challenges, exploiting developments in modeling and pattern recognition from the engineering literature. The net result is the development of an algorithm that safely and effectively delivers insulin to the person with diabetes in order to maintain their blood glucose within a healthy range and avoid short- and long-term complications.

One main objective for our research is immediate clinical application using currently available technology. Artificial pancreas systems using current technology are constrained by the slow action of subcutaneously delivered insulin, as well as delays, lags, and offsets in subcutaneous glucose measurements. To overcome these challenges, we use zone model predictive control (ZMPC). This method uses a model of insulin-glucose interaction to predict the glucose trajectory into the future for a given insulin delivery input. The controller chooses an insulin delivery value that minimizes a cost function that prioritizes safe glucose levels. As an additional safety layer, our artificial pancreas system includes the Health Monitoring System (HMS), which sends escalating alerts for predicted or current hypoglycemia (low blood sugar). Our ZMPC+HMS algorithms have performed well in several clinical studies, and are currently undergoing testing in outpatient clinical trials.

Another objective is to work with collaborators to develop control systems using innovative new strategies for insulin delivery and glucose sensing. New approaches to insulin delivery include the use of fast-acting inhaled insulin for meal boluses or implantable pumps to provide quick insulin action and clearance. A fully implantable artificial pancreas will combine the implantable pump with an implantable sensor to eliminate delays and achieve excellent glycemic control.