Feasibility of Meal Onset Detection Using Electrocardiograms

Information about meal onset is required for the artificial pancreas in efficient blood glucose level control (BGL). In this paper, we explore the effectiveness of electrocardiogram (ECG) signals for the detection of meal onset. In order to detect electrocardiographic changes during the meal ingestion, ECG signals were recorded before, during, and after the meal intake in healthy volunteers using three electrodes in Einthoven’s configuration in four subjects. Six-time domain ECG features were extracted from segments of 20 s duration. These six features were given to a least-squares support vector machine for classification into meal-related ECG and no-meal ECG. The proposed approach achieves a classification accuracy of 84.82%, suggesting that ECG signals carry significant information that can be used to discriminate between meal and non-meal related events. Additionally, the average detection time (from actual meal start to meal onset detection) was only 3.47 min. Our preliminary results suggest that ECG-based meal discrimination information can be used in an artificial pancreas to improve its performance. It may also be used in continuous glucose monitoring systems to remind users of a forgotten meal insulin bolus.