Additional Physical Activity Metrics

Physical activity metrics are algorithms to aggregate raw accelerations signals acquired by acceleration sensors to values that correlate with the intensity of physical activity of a person wearing the acceleration sensor. Theses metrics can be used to estimate energy Expenditure or to directly classify physical activity into activity levels. Most metrics take care of filtering out gravity acceleration and non-movement high frequency acceleration.

Movement Acceleration Intensity (MovementAcceleration)

MAI is a metric based on VanSomeren1996 (opens new window). The acceleration signal is bandpass filtered (Butterworth 0.25-11Hz, 4th order) to remove parts that are not caused by bodily movement. The three axes are fused by the Euclidean norm. The final signal output is the mean value of the output intervall. movisens uses a output intervall of 4 seconds for energy expenditure estimation.

Literature:

• Van Someren, Eus JW, et al. "Gravitational artefact in frequency spectra of movement acceleration: implications for actigraphy in young and elderly subjects." Journal of neuroscience methods 65.1 (1996): 55-62. (opens new window)

Euclidian Norm (PaMetricEn)

The three axes of the 3D acceleration signal are fused to the vector magnitude by computation of the Euclidean norm. The final signal output is the mean value of the output intervall.

Euclidian Norm Minus One (PaMetricEnmo)

Metric ENMO is described in van Hees et al. (2013) (opens new window). The three axes of the 3D acceleration signal are first fused to the vector magnitude by computation of the Euclidean norm. Afterwards the value of gravitational acceleration on earth 1g is subtracted. To inhibit negative subtraction results values smaller than zero are set to zero. The final signal output is the mean value of the output intervall.

Literature:

• Van Hees, Vincent T., et al. "Separating movement and gravity components in an acceleration signal and implications for the assessment of human daily physical activity." PloS one 8.4 (2013): e61691. (opens new window)
• Bakrania, Kishan, et al. "Intensity thresholds on raw acceleration data: Euclidean norm minus one (ENMO) and mean amplitude deviation (MAD) approaches." PloS one 11.10 (2016): e0164045. (opens new window)

High-pass Filtered Euclidean Norm (PaMetricHfen)

Metric HFEN is described in van Hees et al. (2013) (opens new window). The acceleration signal is high-pass filtered (Butterworth 0.2Hz, 4th order) to reduce the effect of gravity on the metric output. The three axes are subsequently fused by computing the Euclidean norm. The final signal output is the mean value of the output intervall.

Literature:

• Van Hees, Vincent T., et al. "Separating movement and gravity components in an acceleration signal and implications for the assessment of human daily physical activity." PloS one 8.4 (2013): e61691. (opens new window)

High-pass Filtered Euclidean Norm plus (PaMetricHfenPlus)

Metric HFEN₊ is described in van Hees et al. (2013) (opens new window). It is based on metric HFEN, which is described in the previous section. The idea behind the signal flow of HFEN₊ is, that the low-pass filtered signal contains the gravitation. The metric output should correlate to the acceleration caused by physical activity. The influence of gravity should be minimized as a consequence. The value of gravitational acceleration on earth 1g is subtracted from the low-pass filtered signal (Butterworth 0.2Hz, 4th order). The result of this subtraction is claimed to include artificial accelerations of low-frequency. So, the result is added to the high-pass filtered signal subsequently after setting negative values to zero. The final signal output is the mean value of the output intervall.

Literature:

• Van Hees, Vincent T., et al. "Separating movement and gravity components in an acceleration signal and implications for the assessment of human daily physical activity." PloS one 8.4 (2013): e61691. (opens new window)

Band-pass Filtered Euclidean Norm (PaMetricBfenMean)

Metric BFEN is described in van Hees et al. (2013) (opens new window). It is similar to metric HFEN, but in addition to high-pass filtering the acceleration signal is low-pass filtered as well, resulting in a bandpass filtered (Butterworth 0.2-15Hz, 4th order) signal. By reducing the content of high frequencies, the influence of noise should be reduced. Subsequently to the filtering process the three axes are fused to the vector magnitude and finally the mean value of the output intervall is calculated.

Literature:

• Van Hees, Vincent T., et al. "Separating movement and gravity components in an acceleration signal and implications for the assessment of human daily physical activity." PloS one 8.4 (2013): e61691. (opens new window)

EEAC (PaMetricEeacMean)

The metric EEAC is described in Härtel et al. (2011) (opens new window). It describes the Euclidean norm of the deviation of the average. First the mean value of a set segment length (suggested by author: 1 second) is calculated and subtracted of the acceleration signal. The Euclidean norm is calculated subsequently. The final signal output is the mean value of the output intervall.

Literature:

• Härtel, Sascha, et al. "Estimation of energy expenditure using accelerometers and activity-based energy models—validation of a new device." European Review of Aging and Physical Activity 8.2 (2011): 109. (opens new window)

Mean Amplitude Deviation (PaMetricMeanAmplitudeDeviation)

The metric MAD is described in Vähä-Ypyä et al. (2015) (opens new window). In the literature it is also called Vector Magnitude Count (VMC) to avoid the confusion between the two MAD acronyms used in accelerometry literature, one for mean amplitude deviation and one for median amplitude deviation. The metric makes use of the mean signal value to reduce the effect of the constant gravitational acceleration.

Literature:

• Vähä-Ypyä, Henri, et al. "Validation of cut-points for evaluating the intensity of physical activity with accelerometry-based mean amplitude deviation (MAD)." PloS one 10.8 (2015): e0134813 (opens new window)
• Vähä‐Ypyä, Henri, et al. "A universal, accurate intensity‐based classification of different physical activities using raw data of accelerometer." Clinical physiology and functional imaging 35.1 (2015): 64-70. (opens new window)
• Bakrania, Kishan, et al. "Intensity thresholds on raw acceleration data: Euclidean norm minus one (ENMO) and mean amplitude deviation (MAD) approaches." PloS one 11.10 (2016): e0164045. (opens new window)
• Karas, Marta, et al. "Accelerometry data in health research: challenges and opportunities." Statistics in Biosciences (2018): 1-28. (opens new window)

Zero Crossing (PaMetricZeroCrossing)

The metric ZC is described by Acebo et al. (1999) (opens new window) and Jean-Louis et al. (2001) (opens new window). ZC is based on just one axis of the acceleration signal because it was designed for a mono-axial accelerometer. The algorithm counts the number of threshold crossings of the band-pass filtered (Butterworth 2-3Hz, 4th order) acceleration signal in an epoch. The downward facing axis of the accelerometer is used.

Literature:

• Acebo, Christine, et al. "Estimating sleep patterns with activity monitoring in children and adolescents: how many nights are necessary for reliable measures?." Sleep 22.1 (1999): 95-103. (opens new window)
• Jean-Louis, Girardin, et al. "Sleep estimation from wrist movement quantified by different actigraphic modalities." Journal of neuroscience methods 105.2 (2001): 185-191. (opens new window)

Proportional Integrating Measure (PaMetricPim)

The metric PIM is introduced by Jean-Louis et al. (2001) (opens new window). Like ZC, which is described in the previous section, PIM is based on just one axis of the acceleration signal because it was designed for a mono-axial accelerometer. The metric output of PIM is based on integration of the surface under the band-pass filtered (Butterworth 2-3Hz, 4th order) acceleration graph. The amplitude of that graph is bound to an upper limit (0.384g). The input of the algorithm is the signal of the downward facing axis of the accelerometer.

Literature:

• Jean-Louis, Girardin, et al. "Sleep estimation from wrist movement quantified by different actigraphic modalities." Journal of neuroscience methods 105.2 (2001): 185-191. (opens new window)

Actiwatch 4 Counts (PaActiWatch4Counts)

The metric Actiwatch 4 Counts is introduced by CamNtech Ltd. Actiwatch 4 Counts are based on just one axis of the acceleration signal because it was designed for a mono-axial accelerometer. The counts are calculated by truncating the acceleration signal to a maximum of 5g. The resulting signal is high-pass (Butterworth 3Hz, 1st order) and low-pass (Butterworth 11Hz, 2nd order) filtered, downsampled to 32 Hz and scaled to values from 0 - 128. Subsequently the maximum in 1 second intervalls is picked and rounded to the nearest integer value. A dead band up to including value 2 is introduced. The final signal output is the sum over the output intervall.

Activity level calculated from physical activity metrics (ActivityLevelFromPaMetric)

You can calculate the activity level from the following paMetrics

• Movement Acceleration Intensity
• Mean Amplitude Deviation
• Euclidian Norm Minus One

You have to specify which physical activity you want to use and the respective cutpoints. This can be done by a custom config file. Here is an example where Movement Acceleration Intensity is used. Please consider that the cutpoints will be dependent on sensor location.

The calculation of the physical activity metric and the activity level is based on the global output interval set in DataAnalyzer.