Estimating an individual’s oxygen uptake during cycling exercise with a recurrent neural network trained from easy-to-obtain inputs: A pilot study

Abstract

Measurement of oxygen uptake during exercise ( _ VO 2 ) is currently non-accessible to most individuals without expensive and invasive equipment. The goal of this pilot study was to estimate cycling _ VO 2 from easy-to-obtain inputs, such as heart rate, mechanical power output, cadence and respiratory frequency. To this end, a recurrent neural network was trained from laboratory cycling data to predict _ VO 2 values. Data were collected on 7 amateur cyclists during a graded exercise test, two arbitrary protocols (Prot-1 and -2) and an "all-out" Wingate test. In Trial-1, a neural network was trained with data from a graded exercise test, Prot-1 and Wingate, before being tested against Prot-2. In Trial-2, a neural network was trained using data from the graded exercise test, Prot-1 and 2, before being tested against the Wingate test. Two analytical models (Models 1 and 2) were used to compare the predictive performance of the neural network. Predictive performance of the neural network was high during both Trial-1 (MAE = 229(35) mlO 2 min -1 , r = 0.94) and Trial-2 (MAE = 304(150) mlO 2 min -1 , r = 0.89). As expected, the predictive ability of Models 1 and 2 deteriorated from Trial-1 to Trial-2. Results suggest that recurrent neural networks have the potential to predict the individual _ VO 2 response from easy-to-obtain inputs across a wide range of cycling intensities.

Temporal convolutional networks predict dynamic oxygen uptake response from wearable sensors across exercise intensities

“…The four cardiorespiratory biosignals derived from the smart shirt (HR, HR reserve, breathing frequency, and minute ventilation ( )) and the work rate (WR) profile were used as inputs into a chain of residual blocks, followed by a dense layer and linear activation to predict at each time point. Results of the TCN networks were compared against a stacked long short-term memory (LSTM) network 26 and random forest (RF) 27 prediction models. The stacked LSTM model was trained using the originally proposed features 23 , as well as adding HR reserve and to the feature set, with a sequence length of 140 s approximating 70 breaths at low intensity exercise.…”

“…Results of the TCN networks were compared against a stacked long short-term memory (LSTM) network 26 and random forest (RF) 27 prediction models. The stacked LSTM model was trained using the originally proposed features 23 , as well as adding HR reserve and to the feature set, with a sequence length of 140 s approximating 70 breaths at low intensity exercise. The RF model was built using the optimal number of trees according to the validation loss (30 trees; see Supplementary Fig.…”

“… Prediction curves are shown for VO2-TCN (proposed) with the lowest validation loss (VO2-TCN A) and 1 s receptive field (VO2-TCN NH), random forest (RF) 21 , and stacked long short-term memory network (LSTM) 23 . Dashed gray line represents the participant’s ventilatory threshold.…”

“…Sensors embedded in textile fabrics have enabled prediction in low-to-moderate-intensity exercise during activities of daily living 21 , 22 . Recent studies have leveraged the time-dependent nature of by modeling the regression as a time-series, or sequential, prediction, where previous physiological states were used to predict 23 and ventilatory threshold (VT) 24 during stationary cycling on a cycle ergometer. Although sequential prediction has been shown to model the temporal nature of oxygen uptake well, further investigations are needed to assess model efficiency and performance across a range of exercise intensities, different days, and demographics.…”

Pop Music Trend and Image Analysis Based on Big Data Technology

“…The ARIMA model can achieve the accurate prediction of data. When the obtained data features are not very stable, the ARIMA model can be used to steady the data features using the initial difference method [ 30 – 32 ].…”