(625d) Event Detection and Estimation of Its Influence Based on Fuzzy Qualitative Representation of Measurements and Fuzzy Logic Estimator

Sediqeh Samadi ^*, Kamuran Turksoy ^**, Iman Hajizadeh ^*, Jianyuan Feng ^*, Mert Sevil ^**, Ali Cinar^*

* Department of Chemical and Biological Engineering, Illinois Institute of Technology, Chicago, IL, USA
(Corresponding author: cinar@iit.edu).
** Department of Biomedical Engineering, Illinois Institute of Technology, Chicago, IL, USA.

Detection of underlying (temporal) events in chemical and biological systems is important in order to respond them by taking the appropriate control actions. The reflection of such events on the measured variables is employed to extract features for detection algorithms [1]. In this study, quantitative change of measured variable(s) over moving specified length time window (episode) is translated to the qualitative variables (shapes). Combination of first and second derivative (or difference) signs of every shape is distinct. We introduce fuzziness in qualitative variables where more than one shape can describe an episode. Fuzzy qualitative variables can be used to create the features for detection of the events. In addition to detection of the events, the estimation of their influence of is crucial for control purpose.

Using the proposed fuzzy qualitative representation of measurements, we define a detection feature named “increase of glucose trend index” to identify the occurrence of the meal. Detection of the meals based on continuously measured glucose is of interest for development of artificial pancreas (AP) control systems for people with type 1 diabetes mellitus (T1DM). The algorithm can be incorporated with the AP systems without manual meal announcements, or as a safety feature for people who may forget entering meal information manually. Using qualitative representation of the continuously measured glucose, a time period labeled as meal flag is identified. The criteria for activation and deactivation of meal flag described based on the numerical value of detection feature (increase of glucose trend index) computed every sampling time [2]. Following the detection of the occurrence of the meal, a fuzzy logic based estimator estimates the carbohydrate (CHO) content of the meal to determine the appropriate dose of insulin bolus for a meal. At every sampling time during active meal flag, a fuzzy system with two proposed input variables estimates the absorbed CHO and computes insulin meal bolus dose accordingly. The first input of fuzzy estimator is the ratio of glucose increase with respect to its increase at the detection time and the second input is the absorbed insulin modeled by a normalized finite impulse response (FIR) function [3].

To assess the performance of the algorithm, we used the academic version of UVa/Padova simulator [4] that contains 30 in silico patients (10 adults, 10 adolescents, 10 children). Basal insulin and CHO to insulin ratio for each patient are equal to their default values in the simulator. The proposed algorithm computes the time and dose of bolus insulin. For the evaluation data set, rates of sensitivity of meal detection are 87%, 94%, and 93% and false detection rates are 21%, 4%, and 3% for adults, adolescents and children, respectively. The absolute error in CHO estimation is 23.1%. The mean glucose concentration and the percentages in target range (70-180 mg/dl) are 135 mg/dl, 147 mg/dl, 145 mg/dl and 86.5%, 75%, 74.8% for adults, adolescents, and children, respectively.

 

[1] B. R. Bakshi and G. Stephanopoulos, “Representation of process trends-III. multiscale extraction of trends from process data,” Comput. Chem. Eng., vol. 18, no. 4, pp. 267–302, 1994.

 

[2] S. Samadi, K. Turksoy, I. Hajizadeh, J. Feng, M. Sevil, and A. Cinar, “Meal Detection and Carbohydrate Estimation Using Continuous Glucose Sensor Data ,” IEEE J. Biomed. Heal. Informatics, doi: 10.1109/JBHI.2017.2677953, 2017.

[3] Q. Wang, P. Molenaar, S. Harsh, K. Freeman, J. Xie, C. Gold, M. Rovine, and J. Ulbrecht, “Personalized state-space modeling of glucose dynamics for Type 1 diabetes using continuously monitored glucose, insulin dose, and meal intake an extended Kalman filter approach,” J. Diabetes Sci. Technol., vol. 8, no. 2, pp. 331–345, 2014.

[4] C. D. Man, F. Micheletto, D. Lv, M. Breton, B. Kovatchev, and C. Cobelli, “The UVA/PADOVA type 1 diabetes simulator: new features.,” J. Diabetes Sci. Technol., vol. 8, no. 1, pp. 26–34, 2014.