Accelerometry-based Energy Expenditure Estimation During Activities of Daily Living: A Comparison Among Different...

Area of Science:

• Biomedical Engineering
• Sports Science
• Wearable Technology

Background:

• Physical Activity Energy Expenditure (PAEE) is vital for obesity management.
• Accelerometers can estimate PAEE from body movements, with body center of mass (COM) acceleration being a strong predictor.
• Wrist-worn devices are popular, but their effectiveness for PAEE estimation needs comparison with COM-based placements.

Purpose of the Study:

• To evaluate and compare the accuracy of PAEE estimation using accelerometer placements near the body center of mass (COM) versus wrist-based placements.
• To assess the performance of different COM-based configurations (pelvis-only vs. pelvis and thighs) and wrist-based configurations (left vs. right wrist).
• To compare two established PAEE estimation methods (linear regression and CNN-LSTM) across these different sensor placements.

Main Methods:

• Collected respiratory data (reference) and accelerometer data from 9 participants during daily activities.
• Implemented linear regression (LR) and CNN-LSTM models to estimate PAEE.
• Evaluated accelerometer placements: pelvis-only, pelvis and two thighs (3-acc) for COM-based settings, and left/right wrist for wrist-based settings.

Main Results:

• COM-based settings, particularly the 3-acc configuration, yielded the best PAEE estimation results (LR: R²=0.41, CNN-LSTM: R²=0.53).
• No significant difference was observed between 3-acc and pelvis-acc settings (p=0.278).
• Wrist-based settings (l-wrist-acc, r-wrist-acc) showed no predictive power for PAEE (R²≈0), significantly underperforming COM-based settings (p<0.05).

Conclusions:

• Accelerometer placement near the body center of mass (COM) is superior for accurately estimating physical activity energy expenditure (PAEE) compared to wrist-based placements.
• The 3-accelerometer configuration (pelvis and thighs) and pelvis-only configuration demonstrate comparable effectiveness for PAEE estimation.
• Current wrist-worn accelerometers are not suitable for reliable PAEE prediction, highlighting the importance of sensor location in wearable health monitoring.