Estimation Of Energy Expenditure From Wrist Attached Accelerometer Signal Using Movement Counting And Filtering

Abstract

Activity monitoring from the wrist has yielded accurate results for estimation of time spent in bed rest and energy expenditure (EE) in walking, but the results have not been equally good in everyday tasks. Overestimation of EE has resulted when hands move a lot while whole body movement is low. Recently, complex multi-sensor measurement and computer approaches with pattern recognition have been tested to solve the problem. We used a simple 1-D accelerometer-based movement counter equipped with a pulse filtering procedure (European Patent 1532924), and exposed our method to selected everyday activities. PURPOSE: To determine the suitability of 1-D accelerometer measurement together with a movement counting and filtering procedure in estimation of EE in everyday activities. METHODS: 8 males (27 ± 5 years, BMI 24.0 ± 3.8 kgm-2, mean ± SD) and 7 females (32 ± 6 years, BMI 21.8 ± 1.5 kgm-2) wore a wrist watch type instrument equipped with 1-D accelerometer in laboratory conditions. The protocol included nine activities with various contributions of hand activity: rest, reading, playing with toys, computer work, hitting nails, packing a bag while standing, cleaning the floor and treadmill walking 3.5 km/h and 5.5 km/h. The instrument was tuned to low frequency movements (0.3-1.5 Hz) and it was programmed to register the movement if acceleration exceeded 0.1 g. Depending on whether the majority of movements were periodic or non-periodic, the filter only allowed one movement in 600 ms or 850 ms. Minute-by-minute movement count was transformed to METs using a non-linear relationship and body height based calibration factor. METs were transformed to kJ/min using WHO equations with age and body weight. Reference EE was measured with portable respiratory gas analysis system. Each activity lasted five minutes and the last three minutes of each activity were selected for analysis. RESULTS: Estimated energy expenditure correlated highly (r=0.88) with measured one. Difference between estimated and measured energy expenditure was 0.5 ± 3.1 kJ/min or -2 ± 27% (mean ± SD). The filtering diminished the overestimation in activities where hand movements dominate. CONCLUSIONS: 1-D accelerometer-based movement counter with simple filtering procedure provides accurate estimations of low-to-moderate intensity EE.