Abstract
Oxygen consumption (dot{,{{mbox{V}}}}{{{mbox{O}}}}_{2}) provides established clinical and physiological indicators of cardiorespiratory function and exercise capacity. However, dot{,{{mbox{V}}}}{{{mbox{O}}}}_{2} monitoring is largely limited to specialized laboratory settings, making its widespread monitoring elusive. Here we investigate temporal prediction of dot{,{{mbox{V}}}}{{{mbox{O}}}}_{2} from wearable sensors during cycle ergometer exercise using a temporal convolutional network (TCN). Cardiorespiratory signals were acquired from a smart shirt with integrated textile sensors alongside ground-truth dot{,{{mbox{V}}}}{{{mbox{O}}}}_{2} from a metabolic system on 22 young healthy adults. Participants performed one ramp-incremental and three pseudorandom binary sequence exercise protocols to assess a range of dot{,{{mbox{V}}}}{{{mbox{O}}}}_{2} dynamics. A TCN model was developed using causal convolutions across an effective history length to model the time-dependent nature of dot{,{{mbox{V}}}}{{{mbox{O}}}}_{2}. Optimal history length was determined through minimum validation loss across hyperparameter values. The best performing model encoded 218 s history length (TCN-VO2 A), with 187, 97, and 76 s yielding <3% deviation from the optimal validation loss. TCN-VO2 A showed strong prediction accuracy (mean, 95% CI) across all exercise intensities (−22 ml min−1, [−262, 218]), spanning transitions from low–moderate (−23 ml min−1, [−250, 204]), low–high (14 ml min−1, [−252, 280]), ventilatory threshold–high (−49 ml min−1, [−274, 176]), and maximal (−32 ml min−1, [−261, 197]) exercise. Second-by-second classification of physical activity across 16,090 s of predicted dot{,{{mbox{V}}}}{{{mbox{O}}}}_{2} was able to discern between vigorous, moderate, and light activity with high accuracy (94.1%). This system enables quantitative aerobic activity monitoring in non-laboratory settings, when combined with tidal volume and heart rate reserve calibration, across a range of exercise intensities using wearable sensors for monitoring exercise prescription adherence and personal fitness.
Highlights
Cardiorespiratory fitness is an established risk factor for cardiovascular disease and all-cause mortality[1] and is an important determinant for endurance exercise performance[2]
We propose and evaluate a sequential deep with larger receptive fields, apart from the two-filter models that learning model based on temporal convolutional networks (TCNs)[25] for predicting V_O2 from physiological inputs derived from smart textiles and a cycle ergometer
The best performing model according to holdout validation loss (TCN-VO2 A) had a 218 s receptive field and 19,921 parameters
Summary
Cardiorespiratory fitness is an established risk factor for cardiovascular disease and all-cause mortality[1] and is an important determinant for endurance exercise performance[2]. Cardiorespiratory fitness is conventionally assessed by measuring the rate of oxygen consumption ( V_ O2) and its dynamic response to exercise. Recent advances in wearable technologies and artificial intelligence have led to new developments in non-intrusive cardiorespiratory monitoring These approaches are generally modeled as regression problems, where a machine learning model learns a transformation function between physiological inputs from wearable sensors and V_ O2 measured using a gas analyzer system. We assessed the effect of receptive field and model complexity on prediction accuracy to investigate the temporal relationship between physiological inputs and V_ O2 response to provide guidance on optimal model design and assessment across a range of different exercise intensities.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
