GlucoSense: Non-Invasive Glucose Monitoring using Mobile Devices

Regular glucose monitoring is crucial for diabetic patients to avoid the risk of health complications such as stroke, kidney failure, heart disease, and even death. Most current devices for measuring glucose are costly and painful. We propose GlucoSense, a non-invasive glucose sensing solution on mo- bile devices. GlucoSense builds on the fact that glucose is an optically active molecule, which interacts with various wave- lengths. We first conduct spectral analysis to demonstrate the feasibility of measuring glucose in the visible and near- infrared range (400–1000 nm), which is the range available on mobile devices. We also identify the relative importance of various spectral bands in this range. We further propose multiple practical designs for obtaining the required spectral bands for measuring glucose. We then design GlucoSense exploiting the sensing capabilities of modern smartphones combined with machine learning models. We conduct an ethics-approved user study with a diverse set of participants in terms of age, sex, ethnicity, and body mass index (BMI). We compare GlucoSense against a widely-used, FDA-approved glucose measuring device. Our results show that 80.4% of Glu- coSense predictions are within Zone A (clinically accurate), and the remaining 19.3% are in Zone B (clinically accept- able) of the Clarke Error Grid (CEG). In addition, 99.7% of the predictions are within the None and Slight risk zones of the Surveillance Error Grid (SEG), indicating their high accuracy. Both CEG and SEG are standard metrics for assess- ing glucose-measuring devices. These results were obtained by GlucoSense running on unmodified phones in realistic environments with diverse illuminations.

In The 31st Annual International Conference on Mobile Computing and Networking