부산대학교 도서관

Continuous glucose monitoring (CGM) sensors are extensively used for diabetes management. These sensors sample glucose from interstitial fluid (ISF) and provide insights into glucose trajectories. Over the years, many different types of CGM sensors involving enzymatic or nonenzymatic methods of sensing have been demonstrated. In CGM sensors, the degradation in sensor sensitivity postimplantation could be either due to the degradation of the glucose oxidase (GOx) enzyme used in enzymatic glucose sensors or due to biofouling. So, the majority of commercially available CGM sensors recommend users manually calibrate their CGM sensors daily. To avoid this, previous autocalibration approaches have focused on mathematical modeling methods to estimate the sensitivity of a sensor. These methods have shown limited functionality in real-life scenarios due to the variability of the environment surrounding the sensor. In this article, we propose an electrochemical impedance spectroscopy (EIS)-driven approach that predicts the sensitivity of the sensor using machine-learning (ML) methods applied to EIS parameters. First, we demonstrate that degradation in glucose sensitivity increases in the presence of biofouling in addition to the reduction in GOx enzyme activity. Biofouling leads to the formation of films on the sensor surface, leading to changes in EIS parameters. The results from our method predict the sensitivity of the electrode with a mean absolute error (MAE) of 1.50 nA/mM in an in vitro setup, using a random forest regression model. This article demonstrates that EIS parameters can be utilized to predict sensitivity in enzymatic glucose sensors.