Neuroinform (2010) 8:197–200

DOI 10.1007/s12021-010-9075-9

The Biopsychology—Nonlinear Analysis Toolbox: A Free,

Open-Source Matlab-Toolbox for the Non-linear Analysis
of Time Series Data
Christian Beste & Tobias Otto & Sven Hoffmann

Published online: 8 June 2010

# Springer Science+Business Media, LLC 2010

Abstract We provide a free, open-source toolbox for non- autocorrelation of the signal (Bagarinao et al. 2008), but
linear time series analyses. The major goal of this project also non-linear measures can be extracted (Stam 2005;
was to provide a toolbox for nonlinear time series analyses Kantz and Schreiber 2004). These non-linear measures
that is easily accessible to a wide range of neuroscientists. have been applied in many contexts within neuroscience.
The toolbox offers modular, powerful and flexible algo- Non-linear time series analysis has beforemost been
rithms embedded in an easy to handle graphical user applied to EEG and MEG of healthy subjects during no-
interface (GUI). The toolbox can be run within the Matlab task resting states, perceptual processing and performance
environment, but also as stand-alone solution without in cognitive tasks and different sleep stages. Many pathologic
reference to a programming environment that is also usable states have been examined as well, ranging from toxic states,
for different PC operating systems (Windows and Linux). epileptic seizures, and psychiatric disorders to Alzheimer’s,
The Biopsychology—Nonlinear Analysis Toolbox and Parkinson’s and Creutzfeldt-Jakob disease (e.g. Stam 2005;
documentation are available freely and open-source from Elger and Lehnertz 1998).
http://biopsynltoolbox.sourceforge.net Despite broad potential interest of non-linear time series
analysis measures in different areas of research, these measures
Keywords Non-linear time series analysis . Time series . are rarely used (Goldberg et al. 2009). Partly, this may be due
Matlab . EEG . MEG . Toolbox to the relatively complex mathematical algorithms allowing
the extraction of parameters that give information about the
Within neuroscience, time series data frequently occur. non-linear dynamical properties of a time series (Zhou et al.
These time series can essentially be of any origin, e.g. 2009). There are several software solutions that provide
electroencephalographic (EEG) data, or magnetoencepha- powerful algorithms to extract parameters describing the non-
lographic (MEG) data (Stam 2005). To describe the linear dynamics of a time series (e.g. the TISEAN package;
dynamics of this time series not only linear measures, http://www.mpipks-dresden.mpg.de/∼tisean/) (Hegger et al.
like the power of the signal, frequency spectrum and 1999). These solutions represent an open source that can be
extended by the community, but they require a broader
C. Beste (*) : T. Otto
background in programming skills to be implemented and
Institute for Cognitive Neuroscience, applied to ones own data. On the other hand, commercial
Department of Biopsychology, Ruhr-Universität Bochum, systems are also available, but they provide a cost intensive
Universitätsstrasse 150, and often inflexible solution with less powerful algorithms
44780 Bochum, Germany
e-mail: [email protected]
that cannot be extended.
To overcome these limitations, we here introduce a
S. Hoffmann (*) Matlab-based toolbox mainly using established algorithms
Leibniz Research Centre for Working from the TISEAN-package1 (Hegger et al. 1999) that offers
Environment and Human Factors,
Ardeystr. 67,
1
44139 Dortmund, Germany A full list of included algorithms as well as their corresponding
e-mail: [email protected] licences are part of the toolbox
198 Neuroinform (2010) 8:197–200

modular, powerful and flexible algorithms embedded in an Windows and Linux. In this compiled version, the codes
easy to handle graphical user interface (GUI). Due to these are not editable, but we will provide regular updates of the
features, the Biopsychology-Nonlinear Analysis Toolbox as compiled version to keep also the user of this version
a free, open-source Matlab toolbox is aimed at users informed about latest developments of the Biopsychology-
interested in applying non-linear analysis techniques with Nonlinear Analysis Toolbox. The Toolbox can be down-
no background in programming as well as the experienced loaded from http://biopsynltoolbox.sourceforge.net. This
programmer with large methodological background. The internet site is also ready for help, communal discussions
project as it is presented here reflects the current state of the and upload of extensions and modifications. In the
toolbox. The toolbox should be viewed as work in progress, following, we give a short overview of the toolbox, its
meaning that we will continue to implement further organization and features.
functions into this toolbox, but also everyone else is The graphical user interface (GUI) of the toolbox is
welcomed to contribute to this toolbox. As such, it is a illustrated in Fig. 1.
community-based project ensuring high quality of code and When starting to work with the toolbox the time series
independence from commercial offers (Rose et al. 2008). has first to be loaded into the toolbox. The button, “select
To grant full communal use and flexibility, all code, files” in the “Choose source files” section of the toolbox
documentation of code and documentation of hardware opens a Windows dialog box where files containing the
are published under a creative commons license (CC GNU data of the time series can be marked and transferred into
GPL by attribution, http://creativecommons.org/licenses/ the toolbox. The ASCII files should be organized in such a
GPL/2.0/). This license allows free use, modification and way that the time series data is organized in columns.
redistribution of the toolbox or documentation as long as These ASCII files must be structured in columns giving
the authors and this article are cited. The same applies to the time-series of each channel. The columns are separated
results that were generated with the toolbox and are by spaces. This allows also multi-channel data (e.g. EEG
published. However, to broaden the usability of the toolbox or MEG data) to be processed with this toolbox. Files once
and further increase cost-control, the toolbox is also loaded into the toolbox can also be removed by marking
available as a compiled version that runs independent from the respective file and clicking the “remove from list”
the Matlab environment for two different platforms: i.e. button.

Fig. 1 The biopsychology—nonlinear analysis toolbox

Neuroinform (2010) 8:197–200 199

Once one or more files have been loaded into the Logfile window denotes the successful loading of a previous-
toolbox several algorithms are offered that can be applied to ly save configuration set and the function that have been
the data. The algorithms range from simple measures of defined in this configuration file are also marked.
variance, kurtosis and skewness of the data in the time Once all algorithms and parameters have been chosen or
series, to more advanced measures of autocorrelation of the modified, applying the different algorithms to a set of data
data set, cross-correlations, entropy, Hurst exponent, corre- only requires pressing the “StartBatch” icon. The “batch
lation sum,—dimension and—entropy, as well as the function” implemented will apply each of the marked
mutual information of the data set, the maximal Lyapunov algorithm to each of the files loaded into the toolbox. For
exponent and surrogate data tests (Schreiber and Schmitz multi-channel data this operation will be repeated for each
2000). For all algorithms that require parameters (e.g. data column separately. The results of these analyses are
autocorrelation, correlational sum,—dimension,—entropy, written into output files in the same folder as the original
mutual information and the maximal Lyapunov exponent) data files. The result files are named similar to the original
“Option” buttons are provided that enable adjustments of data files. For multi-channel data, the results of each
the algorithm parameters. When starting the toolbox, these channel are given within this results file for each channel
options are set to default, which corresponds to the default separately. Applying non-linear algorithms to the data is
setup of the TISEAN functions. The user is free to modify very time-consuming. To allow the fastest possible pro-
each of these functions, but this is on his/her own cessing of the data, especially for workstations having
responsibility! Any changes in the algorithm parameters multi-core processors, the toolbox is equipped with an
are documented in the Logfile window at the bottom of the option that allows demanding as much processor cores as
toolbox. This Logfile window further provides a history of possible. The “Number of parallel threads” allows the user
the entire session since starting the toolbox. Once the user to choose from the pull-down menu how many processor
has adjusted the options of the different algorithms and cores (up to eight) s/he would like to use for data
marked the procedures that s/he would like to apply to the processing. The user is free to choose up to eight processor
data it is possible to save this configuration of functions by cores for data processing, by simply choosing the number
clicking the button “Save configuration” in the “Load and from the pull-down menu. Lastly, we implemented a
save settings” section of the toolbox. From here, a plotting function of the data, which in its current form is
previously saved configuration can also be loaded and applied able to plot the results of the correlation sum,—dimension
to a new data set. The user is ensured that a configuration has and—entropy analysis. These plots can be saved as JPEG
been saved by entry in the Logfile window. Similar, the and as a MATLAB figure.

Box 1:

An example data file can be downloaded together with the toolbox (“BioPsychTest”).

Enclosed to this data file an output file is given denoting the computation of the

correlation dimension using the toolbox. This function can be plotted using the toolbox

and cross-checked with one’s own results, if desired. For the calculation of the

correlation dimension and the Lyapunov exponent the default parameters have been

used. For further quality checks the same analysis was conducted using the original

TISEAN package with the same default parameters. The results are the same.

The toolbox was extensively tested in our laboratories several problems related to the data format and output file.
for many months. In the course of this testing we For example, the implemented functions work even though
extensively debugged several problems. As result we decimal seperators chosen may be wrong (dot instead of
implemented several procedures that automatically retain comma). The toolbox recognizes this and adjusts the
200 Neuroinform (2010) 8:197–200

decimal seperators automatically. Moreover information is Information Sharing Statement

automatically deleted. After applying the algorithms to the
data, the original data files will be renamed with another The Biopsychology—Nonlinear Analysis Toolbox and docu-
extension and the output file is also automatically renamed, mentation are available freely and open-source and can be
corresponding to the file name of the original data. The downloaded from http://biopsynltoolbox.sourceforge.net.
toolbox was tested with larger data samples of multichannel
data (32 channels, 500 Hz recording frequency, 30 min). Acknowledgements This work was supported by a Grant from the
“Research Department Neuroscience” Ruhr-University Bochum and
Moreover, the toolbox has been used in courses with
by a Grant from FoRUM AZ F647-2009 to CB.
undergraduate students interested in non-linear time-series
analysis.
The free and open-source Biopsychology-Nonlinear Anal- References
ysis Toolbox provides user of different disciplines in
neuroscience with an easy to handle, yet fully flexible and Bagarinao, E., Matsuo, K., Nakai, T., & Tanaka, Y. (2008). BAX: a
extendable framework to perform non-linear time series toolbox for the dynamic analysis of functional MRI datasets.
Neuroinformatics, 6(2), 109–115.
analyses in neuroscience. The major aim of this project was
Elger, C. E., & Lehnertz, K. (1998). Seizure prediction by non-linear
to enable a broad range of neuroscientists and young time series analysis of brain electrical activity. European Journal
researchers to use non-linear measures of time series analysis of Neuroscience, 10, 786–789.
in a straightforward manner, ensuring biggest possible Goldberg, D. H., Victor, J. D., Gardner, E. P., & Gardner, D. (2009). Spike
train analysis toolkit: enabling wider application of information-
utilization of computational power. This will enable a broader theoretic techniques to neurophysiology. Neuroinformatics, 7(3),
range of researchers to perform non-linear time series analyses 165–178.
in order to estimate whether non-linear time series analyses Hegger, R., Kantz, A., & Schreiber, T. (1999). Practical implementa-
may advance a specific field of research. While compiling this tion of nonlinear time series models: the TISEAN package.
Chaos, 9, 413.
toolbox, approved algorithms from the TISEAN as well other
Kantz, A., & Schreiber, T. (2004). Nonlinear time series analysis
open-source codes have been implemented by the highest (2nd ed.). Cambridge: Cambridge University Press.
possible quality of the algorithms. By providing a community- Rose, J., Otto, T., & Dittrich, L. (2008). The biopsychology-Toolbox:
based platform we would like to extend the scope of the a free, open-source Matlab-toolbox for the control of behavioral
experiments. Journal of Neuroscience Methods, 175, 104–107.
toolbox. In the near future we will provide versions that can be
Schreiber, T., & Schmitz, A. (2000). Surrogate time series. Physica D,
run under Macintosh, too. The toolbox is given without any 142, 346.
warranty. If users decide to use these programs, they do so Stam, C. J. (2005). Nonlinear dynamical analysis of EEG and MEG:
entirely at their own risk. Further, the toolbox does not review of an emerging field. Clinical Neurophysiology, 116,
2266–2301.
discharge any potential user from familiarization with the
Zhou, S., Wu, Y., & Dos Santos, C. C. (2009). Visualizing dynamical
mathematical and theoretical concepts underlying non-linear neural assemblies with a fuzzy synchronization clustering
time series analyses. analysis. Neuroinformatics, 7(4), 233–244.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.