Limits...
Bump time-frequency toolbox: a toolbox for time-frequency oscillatory bursts extraction in electrophysiological signals.

Vialatte FB, Solé-Casals J, Dauwels J, Maurice M, Cichocki A - BMC Neurosci (2009)

Bottom Line: oscillatory activity, which can be separated in background and oscillatory burst pattern activities, is supposed to be representative of local synchronies of neural assemblies.Oscillatory burst events should consequently play a specific functional role, distinct from background EEG activity - especially for cognitive tasks (e.g. working memory tasks), binding mechanisms and perceptual dynamics (e.g. visual binding), or in clinical contexts (e.g. effects of brain disorders).The software is provided with a Matlab toolbox which can compute wavelet representations before calling automatically the stand-alone application.

View Article: PubMed Central - HTML - PubMed

Affiliation: Riken BSI, Lab, ABSP, Wako-Shi, Japan. fvialatte@brain.riken.jp

ABSTRACT

Background: oscillatory activity, which can be separated in background and oscillatory burst pattern activities, is supposed to be representative of local synchronies of neural assemblies. Oscillatory burst events should consequently play a specific functional role, distinct from background EEG activity - especially for cognitive tasks (e.g. working memory tasks), binding mechanisms and perceptual dynamics (e.g. visual binding), or in clinical contexts (e.g. effects of brain disorders). However extracting oscillatory events in single trials, with a reliable and consistent method, is not a simple task.

Results: in this work we propose a user-friendly stand-alone toolbox, which models in a reasonable time a bump time-frequency model from the wavelet representations of a set of signals. The software is provided with a Matlab toolbox which can compute wavelet representations before calling automatically the stand-alone application.

Conclusion: The tool is publicly available as a freeware at the address: http://www.bsp.brain.riken.jp/bumptoolbox/toolbox_home.html.

Show MeSH

Related in: MedlinePlus

Sparse time-frequency bump modeling of a toy EEG signal. (a) The toy EEG signal (Biosemi system, 2048 Hz sampling rate, 2 sec), recorded in rest condition with eyes closed, is first (b) transformed using complex Morlet wavelets, then (c) the map is z-scored (offset = 1). Sparse time-frequency bump modeling decomposes the z-scored map into a sum (d) of half ellipsoid (c) parametric functions (windows of 4 cycles, pruned to the 8 first bumps).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2690596&req=5

Figure 1: Sparse time-frequency bump modeling of a toy EEG signal. (a) The toy EEG signal (Biosemi system, 2048 Hz sampling rate, 2 sec), recorded in rest condition with eyes closed, is first (b) transformed using complex Morlet wavelets, then (c) the map is z-scored (offset = 1). Sparse time-frequency bump modeling decomposes the z-scored map into a sum (d) of half ellipsoid (c) parametric functions (windows of 4 cycles, pruned to the 8 first bumps).

Mentions: Instead of using average approaches, one could try to analyze directly single trials in the time-frequency plane. However, in the case of time-frequency planes, hundreds of thousands of coefficients are used to represent a signal; and when a large set of signals is to be compared, the complexity of simple graphical matching methods becomes intractable. Analyzing directly this large amount of information leads to complex computations, and either approximate or over-fitted models (this problem is usually termed as the "curse of dimensionality"). We instead advocate a sparsification approach. The main purpose of sparsification approaches is to extract relevant information within redundant data. Sparse time-frequency bump modeling, a 2D extension of the 1D bump modeling described in [14], was developed for this purpose: sparse time-frequency bump modeling extracts the most prominent bursts within a normalized time-frequency map, by modeling them into a sum of parametric functions (see Fig. 1). Bump modeling is however not the only possible sparsification approach. The ridges [15,16] of wavelet maps can be extracted; but while they are sparse, their biological interpretation is not trivial. Multiway analysis [17] allows the simultaneous extraction of multi-dimensional modes, and can be applied to time-frequency representations of electrophysiological signals (e.g. [18]); however it does not allow the independent analysis of transient oscillations. Wavelet packets [16,19] allow the computation of very sparse time-frequency representation (which can be efficient for signal compression, sometimes also for feature extraction); however, because they provide inaccurate time-frequency locations of oscillatory contents, the use of discrete wavelets is not appropriate for signal analysis – especially for electrophysiological data. Finally, the closest method to bump modeling would be matching pursuit [20], which associates a library of functions to a signal – the signal is thereby decomposed into a set of atomic functions, each with specific time-frequency properties (one function could be used to represent a transient oscillation). Other more straightforward attempts were also made at extracting specific atoms of EEG oscillations, like for instance the extraction of narrow band alpha peak epochs [9], or the analysis of a specific time-frequency area (e.g. [21]), but with limitations due to an a priori defined time-frequency area or frequency range [22]. As a comparison, sparse bump modeling allows an automatic broadband modeling of time-frequency atoms, each of them interpreted as transient local activities of neural assemblies (TOS or TOD).


Bump time-frequency toolbox: a toolbox for time-frequency oscillatory bursts extraction in electrophysiological signals.

Vialatte FB, Solé-Casals J, Dauwels J, Maurice M, Cichocki A - BMC Neurosci (2009)

Sparse time-frequency bump modeling of a toy EEG signal. (a) The toy EEG signal (Biosemi system, 2048 Hz sampling rate, 2 sec), recorded in rest condition with eyes closed, is first (b) transformed using complex Morlet wavelets, then (c) the map is z-scored (offset = 1). Sparse time-frequency bump modeling decomposes the z-scored map into a sum (d) of half ellipsoid (c) parametric functions (windows of 4 cycles, pruned to the 8 first bumps).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2690596&req=5

Figure 1: Sparse time-frequency bump modeling of a toy EEG signal. (a) The toy EEG signal (Biosemi system, 2048 Hz sampling rate, 2 sec), recorded in rest condition with eyes closed, is first (b) transformed using complex Morlet wavelets, then (c) the map is z-scored (offset = 1). Sparse time-frequency bump modeling decomposes the z-scored map into a sum (d) of half ellipsoid (c) parametric functions (windows of 4 cycles, pruned to the 8 first bumps).
Mentions: Instead of using average approaches, one could try to analyze directly single trials in the time-frequency plane. However, in the case of time-frequency planes, hundreds of thousands of coefficients are used to represent a signal; and when a large set of signals is to be compared, the complexity of simple graphical matching methods becomes intractable. Analyzing directly this large amount of information leads to complex computations, and either approximate or over-fitted models (this problem is usually termed as the "curse of dimensionality"). We instead advocate a sparsification approach. The main purpose of sparsification approaches is to extract relevant information within redundant data. Sparse time-frequency bump modeling, a 2D extension of the 1D bump modeling described in [14], was developed for this purpose: sparse time-frequency bump modeling extracts the most prominent bursts within a normalized time-frequency map, by modeling them into a sum of parametric functions (see Fig. 1). Bump modeling is however not the only possible sparsification approach. The ridges [15,16] of wavelet maps can be extracted; but while they are sparse, their biological interpretation is not trivial. Multiway analysis [17] allows the simultaneous extraction of multi-dimensional modes, and can be applied to time-frequency representations of electrophysiological signals (e.g. [18]); however it does not allow the independent analysis of transient oscillations. Wavelet packets [16,19] allow the computation of very sparse time-frequency representation (which can be efficient for signal compression, sometimes also for feature extraction); however, because they provide inaccurate time-frequency locations of oscillatory contents, the use of discrete wavelets is not appropriate for signal analysis – especially for electrophysiological data. Finally, the closest method to bump modeling would be matching pursuit [20], which associates a library of functions to a signal – the signal is thereby decomposed into a set of atomic functions, each with specific time-frequency properties (one function could be used to represent a transient oscillation). Other more straightforward attempts were also made at extracting specific atoms of EEG oscillations, like for instance the extraction of narrow band alpha peak epochs [9], or the analysis of a specific time-frequency area (e.g. [21]), but with limitations due to an a priori defined time-frequency area or frequency range [22]. As a comparison, sparse bump modeling allows an automatic broadband modeling of time-frequency atoms, each of them interpreted as transient local activities of neural assemblies (TOS or TOD).

Bottom Line: oscillatory activity, which can be separated in background and oscillatory burst pattern activities, is supposed to be representative of local synchronies of neural assemblies.Oscillatory burst events should consequently play a specific functional role, distinct from background EEG activity - especially for cognitive tasks (e.g. working memory tasks), binding mechanisms and perceptual dynamics (e.g. visual binding), or in clinical contexts (e.g. effects of brain disorders).The software is provided with a Matlab toolbox which can compute wavelet representations before calling automatically the stand-alone application.

View Article: PubMed Central - HTML - PubMed

Affiliation: Riken BSI, Lab, ABSP, Wako-Shi, Japan. fvialatte@brain.riken.jp

ABSTRACT

Background: oscillatory activity, which can be separated in background and oscillatory burst pattern activities, is supposed to be representative of local synchronies of neural assemblies. Oscillatory burst events should consequently play a specific functional role, distinct from background EEG activity - especially for cognitive tasks (e.g. working memory tasks), binding mechanisms and perceptual dynamics (e.g. visual binding), or in clinical contexts (e.g. effects of brain disorders). However extracting oscillatory events in single trials, with a reliable and consistent method, is not a simple task.

Results: in this work we propose a user-friendly stand-alone toolbox, which models in a reasonable time a bump time-frequency model from the wavelet representations of a set of signals. The software is provided with a Matlab toolbox which can compute wavelet representations before calling automatically the stand-alone application.

Conclusion: The tool is publicly available as a freeware at the address: http://www.bsp.brain.riken.jp/bumptoolbox/toolbox_home.html.

Show MeSH
Related in: MedlinePlus