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

Area of Science:

• Cardiorespiratory Physiology
• Wearable Technology
• Machine Learning

Background:

• Oxygen consumption (VO2) is a key indicator of cardiorespiratory function and exercise capacity.
• Current VO2 monitoring is confined to specialized laboratories, limiting widespread application.
• Developing non-laboratory methods for VO2 assessment is crucial for personalized fitness and clinical monitoring.

Purpose of the Study:

• To investigate the temporal prediction of oxygen consumption (VO2) using wearable sensors during cycling exercise.
• To develop and validate a temporal convolutional network (TCN) model for real-time VO2 estimation.
• To assess the accuracy of the TCN model across various exercise intensities and physical activity levels.

Main Methods:

• Acquired cardiorespiratory signals from a smart shirt with integrated textile sensors during cycle ergometer exercise.
• Utilized a temporal convolutional network (TCN) model incorporating causal convolutions and an effective history length for VO2 prediction.
• Determined optimal model hyperparameters through minimum validation loss and validated performance across different exercise protocols and intensities.

Main Results:

• The best TCN model (TCN-VO2 A) utilized a 218-second history length and demonstrated strong prediction accuracy across all exercise intensities.
• The system achieved high accuracy (94.1%) in classifying physical activity levels (vigorous, moderate, light) based on predicted VO2.
• Predicted VO2 showed minimal deviation (<3%) from optimal validation loss with history lengths of 187, 97, and 76 seconds.

Conclusions:

• A TCN model utilizing wearable sensors can accurately predict oxygen consumption (VO2) during exercise in non-laboratory settings.
• This technology enables quantitative aerobic activity monitoring, aiding in the assessment of exercise prescription adherence and personal fitness.
• The developed system holds promise for broader applications in remote patient monitoring and personalized exercise guidance.