Advanced

Generating ActiGraph Counts from Raw Acceleration Recorded by an Alternative Monitor

Brønd, Jan Christian; Andersen, Lars Bo and Arvidsson, Daniel LU (2017) In Medicine and Science in Sports and Exercise 49(11). p.2351-2360
Abstract

PURPOSE: To implement an aggregation method in Matlab for generating ActiGraph counts from raw acceleration recorded with an alternative accelerometer device and to investigate the validity of the method. METHODS: The aggregation method including the frequency band-pass filter was implemented and optimized based on standardized sinusoidal acceleration signals generated in Matlab and processed in the ActiLife software. Evaluating the validity of the aggregation method was approached using a mechanical setup and with a 24-hour free-living recording using a convenient sample of nine subjects. Counts generated with the aggregation method applied to Axivity AX3 raw acceleration data were compared to counts generated with ActiLife from... (More)

PURPOSE: To implement an aggregation method in Matlab for generating ActiGraph counts from raw acceleration recorded with an alternative accelerometer device and to investigate the validity of the method. METHODS: The aggregation method including the frequency band-pass filter was implemented and optimized based on standardized sinusoidal acceleration signals generated in Matlab and processed in the ActiLife software. Evaluating the validity of the aggregation method was approached using a mechanical setup and with a 24-hour free-living recording using a convenient sample of nine subjects. Counts generated with the aggregation method applied to Axivity AX3 raw acceleration data were compared to counts generated with ActiLife from ActiGraph GT3X+ data. RESULTS: An optimal band-pass filter was fitted resulting in a root mean squared error (RMSE) of 25.7 counts per 10 second and mean absolute error (MAE) of 15.0 counts per second across the full frequency range. The mechanical evaluation of the proposed aggregation method resulted in an absolute mean (sd) difference of -0.11 (0.97) counts per 10 second across all rotational frequencies compared to the original ActiGraph method. Applying the aggregation method to the 24-hour free-living recordings resulted in an epoch level bias ranging from -16.2 to 0.9 counts per 10 second, a relative difference in the averaged physical activity (counts per minute) ranging from -0.5% to 4.7% with a group mean (sd) of 2.2% (1.7%) and a Cohen’s Kappa of 0.945 indicating almost a perfect agreement in the intensity classification. CONCLUSION: The proposed band-pass filter and aggregation method is highly valid for generating ActiGraph counts from raw acceleration data recorded with alternative devices. It would facilitate comparability between studies using different devices collecting raw acceleration data.

(Less)
Please use this url to cite or link to this publication:
author
organization
publishing date
type
Contribution to journal
publication status
published
subject
in
Medicine and Science in Sports and Exercise
volume
49
issue
11
pages
2351 - 2360
publisher
Lippincott Williams & Wilkins
external identifiers
  • scopus:85020735838
  • wos:000413405400025
ISSN
0195-9131
DOI
10.1249/MSS.0000000000001344
language
English
LU publication?
yes
id
7bb545da-ef69-4a66-af1a-dd903e6bb2ac
date added to LUP
2017-08-11 16:01:40
date last changed
2018-11-04 04:31:13
@article{7bb545da-ef69-4a66-af1a-dd903e6bb2ac,
  abstract     = {<p>PURPOSE: To implement an aggregation method in Matlab for generating ActiGraph counts from raw acceleration recorded with an alternative accelerometer device and to investigate the validity of the method. METHODS: The aggregation method including the frequency band-pass filter was implemented and optimized based on standardized sinusoidal acceleration signals generated in Matlab and processed in the ActiLife software. Evaluating the validity of the aggregation method was approached using a mechanical setup and with a 24-hour free-living recording using a convenient sample of nine subjects. Counts generated with the aggregation method applied to Axivity AX3 raw acceleration data were compared to counts generated with ActiLife from ActiGraph GT3X+ data. RESULTS: An optimal band-pass filter was fitted resulting in a root mean squared error (RMSE) of 25.7 counts per 10 second and mean absolute error (MAE) of 15.0 counts per second across the full frequency range. The mechanical evaluation of the proposed aggregation method resulted in an absolute mean (sd) difference of -0.11 (0.97) counts per 10 second across all rotational frequencies compared to the original ActiGraph method. Applying the aggregation method to the 24-hour free-living recordings resulted in an epoch level bias ranging from -16.2 to 0.9 counts per 10 second, a relative difference in the averaged physical activity (counts per minute) ranging from -0.5% to 4.7% with a group mean (sd) of 2.2% (1.7%) and a Cohen’s Kappa of 0.945 indicating almost a perfect agreement in the intensity classification. CONCLUSION: The proposed band-pass filter and aggregation method is highly valid for generating ActiGraph counts from raw acceleration data recorded with alternative devices. It would facilitate comparability between studies using different devices collecting raw acceleration data.</p>},
  author       = {Brønd, Jan Christian and Andersen, Lars Bo and Arvidsson, Daniel},
  issn         = {0195-9131},
  language     = {eng},
  month        = {06},
  number       = {11},
  pages        = {2351--2360},
  publisher    = {Lippincott Williams & Wilkins},
  series       = {Medicine and Science in Sports and Exercise},
  title        = {Generating ActiGraph Counts from Raw Acceleration Recorded by an Alternative Monitor},
  url          = {http://dx.doi.org/10.1249/MSS.0000000000001344},
  volume       = {49},
  year         = {2017},
}