A time-efficient continuous ramp protocol for data-driven walking energy expenditure estimation across multiple...

Area of Science:

• Biomechanics and Exercise Physiology
• Wearable Technology and Machine Learning

Background:

• Estimating walking energy expenditure using wearable devices is crucial for data-driven models.
• Traditional discrete step protocols are time-inefficient for creating diverse datasets needed for deep learning.
• Continuous protocols offer a time-efficient alternative for data collection.

Purpose of the Study:

• To compare the effectiveness of a continuous ramp protocol dataset versus a discrete step protocol dataset for estimating walking energy expenditure.
• To evaluate the performance of deep learning models trained on data from both protocols.

Main Methods:

• Fourteen subjects walked on a treadmill with IMUs, measuring energy expenditure via indirect calorimetry.
• A continuous ramp protocol (1.0-1.75 m/s) and a discrete step protocol were employed.
• Deep learning models were trained on datasets from both protocols and compared with a commercial smartwatch.

Main Results:

• No significant differences in energy expenditure were found between the continuous and discrete protocols after respiratory delay compensation.
• Deep learning models trained on continuous data (10.7% error) outperformed those trained on discrete data (13.1% error) and the smartwatch.
• A single IMU with continuous data enabled low error across a broad speed range.

Conclusions:

• The continuous ramp protocol is a time-efficient method for generating valid walking energy expenditure datasets.
• Richer data diversity from continuous protocols enhances deep learning model performance.
• This approach can be extended to various exercise intensities and locomotion types, potentially replacing traditional indirect calorimetry.