Actant - Activity Analysis Toolbox
Copyright (C) 2013-2014, Maxim Osipov <maxim.osipov@gmail.com>
Copyright (C) 2013-2014, Bart te Lindert <b.te.lindert@nin.knaw.nl>
This toolbox include multiple data analysis algorithms implemented in
Matlab and released under Open Source licences. Respective licences are
referenced directly in the code of specific Matlab functions.
0. About
========
Actant provides functionality for visualization and analysis of behavioral
and environmental timeseries, acquired using Actiwatch-L and GENEActiv
accelerometers as well as Actopsy mobile application. Currently implemented
analysis methods include:
- Non-parametric activity analysis:
Van Someren EJW, et al. "Bright light therapy: improved sensitivity
to its effects on rest-activity rhythms in Alzheimer patients by
application of nonparametric methods." Chronobiology international 1999;
16(4):505-518.
- Sleep analysis:
Oakley NR. "Validation with polysomnography of the Sleepwatch
sleep/wake scoring algorithm used by the Actiwatch activity
monitor system: Technical Report to Mini-Mitter Co., Inc., 1997.
Kushida CA, Chang A, Gadkary C, Guilleminault C, Carrillo O, Dement
WC. "Comparison of actigraphic, polysomnographic, and subjective assessment
of sleep parameters in sleep-disordered patients." Sleep Medicine
2001; 2:389-96.
te Lindert BHW, Van Someren EJW. "Sleep estimates using microelectro-
mechanical systems (MEMS)." SLEEP 2013;36(5):781-789
- Sample Entropy:
Richman, Joshua S, Randall Moorman J. "Physiological time-series analysis
using approximate entropy and sample entropy." American Journal of
Physiology-Heart and Circulatory Physiology 2000; 278(6):H2039-H2049.
- Multiscale Entropy:
Costa, Madalena, Goldberger AL, Peng CK.
"Multiscale entropy analysis of complex physiologic time series."
Physical review letters 2002; 89(6):068102.
- Activity segmentation based on Bayesian changepoint detection:
Adams, R. P., & MacKay, D. J. C. (2007). Bayesian Online Changepoint
Detection. Cambridge, UK.
The toolbox provides flexible interface for integration of both new data
formats and analysis methods.
1. Data formats
===============
Actant reads data in Actiwatch, GENEActiv, Actopsy app (CSV) and Actant
own (MAT) formats. Internally, all datasets represented as Matlab
timeseries objects with the following names:
ACC - 3D accelerometry data (m/s^2 or g)
ACT - activity data (counts or m/s^2 or g)
LIGHT - light data (lux)
TEMP - temperature data (degC)
BUTTON - button press indicators (binary)
SPEED - distance travelled (km/h, inaccurate depending on location)
TEXTS - sms messages (days, each character counted as second)
CALLS - phone calls (days)
Plus, in future:
GYRO - 3D gyroscope (deg)
HUM - humidity (%)
PRES - atmospheric pressure (mbar)
HR - heart rate (bpm)
Other timeseries are possible with specific measures, for example mood
ratings from questionnaires.
Data load methods have the following interface:
function ts = load_format(file)
Arguments:
file - File name
Results:
ts - Structure of time series
Additionally, markup timeseries can be generated using analysis algorithms,
where time vector represents starting points of marked intervals and data
vector represents end points (as datenum).
<ANY> - markup data (days)
For unification of data analysis, all high-dimensional and high-sampling
rate activity data can be converted to epochs, either Actiwatch compatible
(counts) or as average acceleration per epoch (m/s^2).
Actant MAT format includes Matlab variables to initialize Actant internal
state, including datasets, analysis algorithm and results:
actant_datasets - cell array of timeseries objects for datasets
actant_sources - cell array of sources (files or analysis) for 'datasets'
actant_plot - values for display characteristics
actant_analysis
method - analysis method name (matlab function name)
args - cell array method arguments
results - cell array results of analysis (except timeseries)
2. Analysis methods
===================
Analysis methods may generate scalar values and timeseries (with values or
data markup). For example, non-linear activity analysis will generate L5,
M10, RA, IS and IV values plus timeseries with segmentations of L5 and M10.
Windowed spectral analysis will generate timeseries with spectral content
of analyzed signal.
The API of analysis method is as following:
function [ts vals] = actant_analysis(args)
Arguments:
args - Cell array of input timeseries and arguments
Results (all optional):
ts - Cell array of timeseries
vals - Cell array of results
When method called without arguments, array of function arguments and
default values is returned in vals, where the first element of array is '_'
with the name of the analysis method as value.
If argument name is preceeded with "ts_" - it is considered as an index in
the table of timeseries and timeseries object is passed to the analysis
function, in all other argument is passed as a string.
3. Visualization
================
The main plot is N days activity plot and can be overlapped with additional
plot (YY axis, result of windowed analysis) and with markup (transparent
patches, result of segmentation).
The plot is configurable with a number of plots to display, number of days
per plot and overlap between plots (in days), so that both single day plots
can be displayed with N-1-0 and 48 hours actogram with N-2-1 settings.
4. Contributions
================
Authors would appreciate your contributions to the Actant project (code can
be found at https://github.com/maximosipov/actant). Also, you are invited
to contrubute your datasets to Physionet (http://www.physionet.org/).