Article Preview
TopIntroduction
With a 9.3% of the US population suffering from diabetes (10% have Type 1 diabetes (T1D) and need insulin injections), research on blood glucose level management and insulin control algorithms is rapidly gaining momentum (Centers for Disease Control and Prevention, 2014). Currently, multiple insulin pumps exist wherein patients manually adjust the insulin amount needed to keep glucose levels in desired range. However, a commercial automatic artificial pancreas is not available. An Artificial Pancreas (AP) system (Turksoy, Quinn, Littlejohn, & Cinar, 2014a; Turksoy et al., 2014b; Turksoy et al., 2014c; Cobelli, Renard & Kovatchev, 2011) can predict the glucose level evolution in blood and infuse the right amount of insulin in response to that level, when needed. One critical component of the AP is the control algorithm that computes the amount of insulin to be infused at a specific time. Too much insulin will result in hypoglycemia, whereas not enough insulin will cause hyperglycemia. Hypoglycemia (blood Glucose Concentration (GC) <70 mg/dl) can lead to dizziness, fainting, diabetic coma or death. Acute hyperglycemia (GC>250 mg/dl) can lead to ketoacidosis in short term, and blood vessel damage in long term, causing complications such as cardiovascular diseases, neuropathy, blindness, wound ulcers, and kidney failure.
With an open loop algorithm as in (Farmer, Jr., Edgar, & Peppas, 2008), continuous assessment and decision making by the patient or care provider is necessary to adjust the insulin dose. A closed-loop control algorithm, can provide a real autonomous system. With drastic variations that can come from something as simple as eating or exercising, the control algorithm needs a way to anticipate them. One proposed solution given in (Thabit & Hovorka, 2012) is the practice of announcements, every time one of the destabilizing activities is about to take place. Nevertheless, with the constant interaction of the user, the automatic operation will cease.
In this work, we opt for another innovative scheme, a fusion of sensors (Elmenreich, 2007) to implement a multi-variable adaptive identification and control system introduced in (Turksoy et al., 2014c). The control algorithm will be able to predict rapid changes in glucose levels even when exercising. We incorporate fitness tracking sensors (such as heart rate, accelerometer, galvanic skin response or temperature sensors) as well as the glucose sensor data to the AP control algorithm.
The outline of the paper is as follows: Section 2 discusses relevant work in AP systems. Section 3 provides the overview of the proposed platform and introduces the system components. Section 4 describes the communication protocols and the data acquisition using the smartphone. Section 5 presents the database interface for both local and remote data storage, and an interface for the AP control algorithm is shown in Section 6.