Samuel R. LaMunion

Estimating Energy Expenditure with ActiGraph GT9X Inertial Measurement Unit

Purpose The purpose of this study was to explore whether gyroscope and magnetometer data from the ActiGraph GT9X improved accelerometer-based predictions of energy expenditure (EE). Methods Thirty participants (mean ± SD: age, 23.0 ± 2.3 yr; body mass index, 25.2 ± 3.9 kg·m−2) volunteered to complete the study. Participants wore five GT9X monitors (right hip, both wrists, and both ankles) while performing 10 activities ranging from rest to running. A Cosmed K4b2 was worn during the trial, as a criterion measure of EE (30-s averages) expressed in METs. Triaxial accelerometer data (80 Hz) were converted to milli-G using Euclidean norm minus one (ENMO; 1-s epochs). Gyroscope data (100 Hz) were expressed as a vector magnitude (GVM) in degrees per second (1-s epochs) and magnetometer data (100 Hz) were expressed as direction changes per 5 s. Minutes 4–6 of each activity were used for analysis. Three two-regression algorithms were developed for each wear location: 1) ENMO, 2) ENMO and GVM, and 3) ENMO, GVM, and direction changes. Leave-one-participant-out cross-validation was used to evaluate the root mean square error (RMSE) and mean absolute percent error (MAPE) of each algorithm. Results Adding gyroscope to accelerometer-only algorithms resulted in RMSE reductions between 0.0 METs (right wrist) and 0.17 METs (right ankle), and MAPE reductions between 0.1% (right wrist) and 6.0% (hip). When direction changes were added, RMSE changed by ⩽0.03 METs and MAPE by ⩽0.21%. Conclusions The combined use of gyroscope and accelerometer at the hip and ankles improved individual-level prediction of EE compared with accelerometer only. For the wrists, adding gyroscope produced negligible changes. The magnetometer did not meaningfully improve estimates for any algorithms.

Domain agnostic online semantic segmentation for multi-dimensional time series

AbstractUnsupervised semantic segmentation in the time series domain is a much studied problem due to its potential to detect unexpected regularities and regimes in poorly understood data. However, the current techniques have several shortcomings, which have limited the adoption of time series semantic segmentation beyond academic settings for four primary reasons. First, most methods require setting/learning many parameters and thus may have problems generalizing to novel situations. Second, most methods implicitly assume that all the data is segmentable and have difficulty when that assumption is unwarranted. Thirdly, many algorithms are only defined for the single dimensional case, despite the ubiquity of multi-dimensional data. Finally, most research efforts have been confined to the batch case, but online segmentation is clearly more useful and actionable. To address these issues, we present a multi-dimensional algorithm, which is domain agnostic, has only one, easily-determined parameter, and can handle data streaming at a high rate. In this context, we test the algorithm on the largest and most diverse collection of time series datasets ever considered for this task and demonstrate the algorithm’s superiority over current solutions.