Search for: All records

In this research, we examine the potential of measuring physiological variables, including heart rate (HR) and respiration rate (RR) on the upper arm using a wireless multimodal sensing system consisting of an accelerometer, a gyroscope, a three-wavelength photoplethysmography (PPG), single-sided electrocardiography (SS-ECG), and bioimpedance (BioZ). The study included collecting HR data when the subject was at rest and typing, and RR data when the subject was at rest. The data from three wavelengths of PPG and BioZ were collected and compared to the SS-ECG as the standard. The accelerometer and gyro signals were used to exclude data with excessive noise due to motion. The results showed that when the subject remained sedentary, the mean absolute error (MAE) for the HR calculation for all three wavelengths of the PPG modality was less than two bpm, while the BioZ was 3.5 bpm compared with SS-ECG HR. The MAE for typing increased for both modalities and was less than three bpm for all three wavelengths of the PPG but increased to 7.5 bpm for the BioZ. Regarding RR, both modalities resulted in RR within one breath per minute of the SS-ECG modality for the one breathing rate. Overall, all modalities on this upper arm wearable worked well when the subject was sedentary. Still, the SS-ECG and PPG showed less variability for the HR signal in the presence of motion during micro-motions such as typing. 

We present a multitask learning approach to the problem of hypoglycemia (HG) prediction in diabetes. The approach is based on a state-of-the-art time series forecasting model, N-BEATS, and extends it by adding a classification task so that the model performs both glucose forecasting (i.e., predicting future glucose values) and HG prediction (i.e., probability of future HG events sometime within the prediction horizon). We also propose an alternative loss function that penalizes forecasting errors in the HG range. We evaluate the approach on a dataset containing over 1.6M recordings from 112 patients with type 1 diabetes who wore a continuous glucose monitor (CGM) for 90 days. Our results show that the classification branch significantly outperforms the forecasting branch on the problem of HG prediction, and that the new loss function is more effective at reducing forecasting errors in the HG range than multi-task learning.