A Mobile Health Platform for Automated Diet Monitoring Using Continuous Glucose Monitors and Context-Aware Machine Learning

Abstract

Automated diet monitoring, an important tool in preventing healthy individuals and those with pre-diabetes from developing Type 2 Diabetes, requires automatic eating detection and estimation of the macronutrient contents of ingested food. While signals from continuous glucose monitors may track the post-prandial glucose response (glucose response after eating) and use this for estimation of nutritional information, the proper identification and segmentation of these periods of eating require additional sensing modalities and contextual information. In this work, we developed a framework for machine learning modeling to detect eating periods, properly segment post-prandial glucose responses, and estimate nutritional content from these segments in real-world environments using data captured from a continuous glucose monitor and augmented with con-textual data from smartwatch wearable sensors. Using a custom-developed platform, we conduct a human subject study where participants were free to eat what they wished, when they wished, logging data and wearing a set of sensors. To aid future, just-in-time diet monitoring applications, we found that contextual data improved eating moment detection and thus enables real-time macronutrient estimation.