Estimation of the spatiotemporal structure of event-related desynchronization and synchronization in magnetoencephalography.

We present a comprehensive methodology for identifying cerebral areas involved in event-related changes of electromagnetic activity of the human brain, and also for tracing the temporal evolution of this activity. Information from pre- and peristimulus time intervals--in terms of event-related synchronization (ERS) and desynchronization (ERD) of the magnetoencephalographic (MEG) signal--was directly incorporated in the relevant test statistics. For the individual steps of the analysis, we used particular estimations of the time-frequency distribution of the energy along with particular error control methods, that is, short-time Fourier transform and false-discovery rate at the sensor level and multitapers and familywise error rate at the source level. This procedure was applied to two types of group-level tests, a within-condition test and a between-conditions test. The performance of the proposed methodology is assessed by (1) analyzing the event-related brain activity from two experimental conditions of an auditory MEG experiment--passive listening to a sequence of frequency-modulated sweeps and their active categorization with respect to the direction of frequency modulation, and (2) comparing the findings with those obtained with a widely used cluster-based analysis.

Single-trial reconstruction of auditory evoked magnetic fields by means of Template Matching Pursuit.

We present a new paradigm for the adaptive estimation of evoked brain responses in single trials, based upon the combination of the matching pursuit (MP) algorithm and template matching, and referred to as Template Matching Pursuit (TMP). In contrast to the classical template matching with invariant single-trial morphology and to previous approaches using MP with strong similarity constraint on functions in sequential trials, this adaptive approach allows for a wide variety of waveforms, and its universality is retained by parametrizing all relevant waveforms in terms of Gabor functions. A survey of single-trial estimates obtained for 10 subjects (∼4000 individual trials in total) confirms the validity of the assumption of a good approximation of single-trial waveforms. Owing to the fully parametric approach, we can easily perform also any quantitative analysis of such a huge dataset. As an example we take the trial-to-trial variability of the peak amplitude and latency of the auditory M100 component. This methodology provides estimates of diversified morphologies, which makes it free from the limitations inherent to any restrictive model. This seems advantageous in the context of the ongoing debate as to the neural mechanisms of average evoked brain responses.

Information infrastructure for cooperative research in neuroscience.

The paper describes a framework for efficient sharing of knowledge between research groups, which have been working for several years without flaws. The obstacles in cooperation are connected primarily with the lack of platforms for effective exchange of experimental data, models, and algorithms. The solution to these problems is proposed by construction of the platform (EEG.pl) with the semantic aware search scheme between portals. The above approach implanted in the international cooperative projects like NEUROMATH may bring the significant progress in designing efficient methods for neuroscience research.

On the robust parametric detection of EEG artifacts in polysomnographic recordings.

We present an open, parametric system for automatic detection of EEG artifacts in polysomnographic recordings. It relies on independent parameters reflecting the relative presence of each of the eight types of artifacts in a given epoch. An artifact is marked if any of these parameters exceeds a threshold. These thresholds, set for each parameter separately, can be adjusted via "learning by example" procedure (multidimensional minimization with computationally intensive cost function), which can be used to automatically tune the parameters to new types of datasets, environments or requirements. Performance of the system, evaluated on 103 overnight polysomnographic recordings, revealed concordance with decisions of human experts close to the inter-expert agreement. To make this statement well defined, we review the methodology of evaluation for this kind of detection systems. Complete source code is available from http://eeg.pl; a user-friendly version with Java interface is available from http://signalml.org.

Event-related desynchronization and synchronization in MEG: Framework for analysis and illustrative datasets related to discrimination of frequency-modulated tones.

We introduce a complete framework for the calculation of statistically significant event-related desynchronization and synchronization (ERD/ERS) in the time-frequency plane for magnetoencephalographic (MEG) data, and provide free Internet access to software and illustrative datasets related to a classification task of frequency-modulated (FM) tones. Event-related changes in MEG were analysed on the basis of the normal component of the magnetic field acquired by the 148 magnetometers of the hardware configuration of our whole-head MEG device, and by computing planar gradients in longitudinal and latitudinal direction. Time-frequency energy density for the magnetometer as well as the two gradient configurations is first approximated using short-time Fourier transform. Subsequently, detailed information is obtained from high-resolution time-frequency maps for the most interesting sensors by means of the computationally much more demanding matching pursuit parametrization. We argue that the ERD/ERS maps are easier to interpret in the gradient approaches and discuss the superior resolution of the matching pursuit time-frequency representation compared to short-time Fourier and wavelet transforms. Experimental results are accompanied by the following resources, available from http://brain.fuw.edu.pl/MEG: (a) 48 high-resolution figures presenting the results of four subjects in all applicable settings, (b) raw datasets, and (c) complete software environment, allowing to recompute these figures from the raw datasets.

Time-frequency microstructure and statistical significance of ERD and ERS.

ERD and ERS were introduced as the time courses of the average changes of energy in given frequency bands. These curves are naturally embedded in the time-frequency plane. Time-frequency density of signals energy can be estimated by means of a variety of transforms. In general, resolution of these methods depends on a priori choices of parameters regulating the tradeoff between the time and frequency resolutions. As an exception, adaptive time-frequency approximations adapt resolution to the local structures of the analyzed signal. Matching pursuit (MP) algorithm is a reliable implementation of this approach. Its application to the event-related EEG allows for a detailed presentation of the time-frequency microstructure of changes of the average energy density, as well as calculation of high-resolution maps of ERD/ERS in the time-frequency plane. However, even with such a detailed picture of the signal energy changes, their significance remains an open issue. Owing to a stochastic character of the EEG, a visible increase or decrease of energy can occur due to a pure chance or a phenomenon unrelated to the event. For a proper estimation of the statistical significance of ERD/ERS, that is, the average changes of signals energy density in relation to the reference period, we must take into account possibly non-normal distributions of energy, and, especially, the problem of multiple comparisons appearing in hypotheses related to different frequency bands and time epochs. This chapter presents and discusses a complete framework for high-resolution estimation of the ERD/ERS microstructure in the time-frequency regions, revealing statistically significant changes.

Estimation of short-time cross-correlation between frequency bands of event related EEG.

Simultaneous variations of the event-related power changes (ERD/ERS) are often observed in a number of frequency bands. ERD/ERS measures are usually based on the relative changes of power in a given single frequency band. Within such an approach one cannot answer questions concerning the mutual relations between the band-power variations observed in different frequency bands. This paper addresses the problem of estimating and assessing the significance of the average cross-correlation between ERD/ERS phenomena occurring in two frequency bands. The cross-correlation function in a natural way also provides estimation of the delay between ERD/ERS in those bands. The proposed method is based on estimating the short-time cross-correlation function between relative changes of power in two selected frequency bands. The cross-correlation function is estimated in each trial separately and then averaged across trials. The significance of those mean cross-correlation functions is evaluated by means of a nonparametric test. The basic properties of the method are presented on simulated signals, and an example application to real EEG and ECoG signals is given.

Efficient application of Internet databases for new signal processing methods.

This paper highlights the ways in which Internet databases may be efficiently used to foster the application of progress in biomedical sciences via data sharing and new algorithms. Employing the Internet to accelerate the pace of interdisciplinary research has significant potential, yet as with all new technologies, the first applications often cause more disappointment than positive outcomes. We discuss examples of solutions to the basic issues: (1) finding the relevant datasets (in portals connected via the Inter-neuro infrastructure), (2) reading the particular format in which the data was stored (using the SignalML language for metadescription of time series), (3) choosing the right method for the data analysis (we provide a brief review of the methods used for the analysis of EEGs, and discuss two of them in detail: Directed Transfer Function and Matching Pursuit), and (4) sharing the software for chosen methods of analysis (via repositories such as the eeg.pl thematic portal).

Computationally efficient approaches to calculating significant ERD/ERS changes in the time-frequency plane.

This paper addresses some practical issues related to the calculation, display and assessment of the significance of changes in the average time-frequency energy density of event-related brain activity. Using scalp EEG and subdural ECoG example datasets, parametric tests are evaluated as a replacement for previously applied computer-intensive resampling methods. The performance of different estimates of energy density, based on matching pursuit, scalogram and spectrogram, and their Box-Cox transformations is evaluated with respect to the assumption of normality required for the t-test, and the consistency of the final results.

Propagation of EEG activity in the beta and gamma band during movement imagery in humans.

OBJECTIVES: The objective of the paper was the determination of electrical brain activity propagation in sensorimotor areas during hand movement imagery. METHODS: Right-hand and left-hand movement imagination was studied in three subjects. The 10-channel Multivariate Autoregressive Model (MVAR) was fitted to EEG signals recorded from subsets of electrodes overlying central and related brain areas. By means of the Short-time Directed Transfer Function (SDTF) the propagation of brain activity as a function of frequency and time was found. RESULTS: During imagery the relation between propagations in gamma and beta bands changed significantly for electrodes overlying sensorimotor areas, namely the increase in gamma was accompanied by the decrease in the beta band. CONCLUSIONS: The hypothesis was put forward that these kinds of changes in flow of electrical brain activity are connected with the specific information processing.

Time-frequency analysis of brain electrical activity--adaptive approximations.

OBJECTIVES: We present an approach to time-frequency analysis of bioelectrical signals. METHODS: The method relays on the decomposition of the signal into a set of waveforms that have good localization both in time and in frequency. The waveforms belong to a highly redundant set of functions - allowing for a very accurate description of signal components. RESULTS: Properties of the method are illustrated by simulations and applications to EEG. CONCLUSION: The presented method delivers a common formalism suitable for describing both gross statistical properties of structures present in bioelectrical signals, as well as microstructure of chosen phenomena.

A simple system for detection of EEG artifacts in polysomnographic recordings.

We present an efficient parametric system for automatic detection of electroencephalogram (EEG) artifacts in polysomnographic recordings. For each of the selected types of artifacts, a relevant parameter was calculated for a given epoch. If any of these parameters exceeded a threshold, the epoch was marked as an artifact. Performance of the system, evaluated on 18 overnight polysomnographic recordings, revealed concordance with decisions of human experts close to the interexpert agreement and the repeatability of expert's decisions, assessed via a double-blind test. Complete software (Matlab source code) for the presented system is freely available from the Internet at http://brain.fuw.edu.pl/artifacts.

Unbiased high resolution method of EEG analysis in time-frequency space.

Matching Pursuit (MP)--a method of high-resolution signal analysis--is described in the context of other methods operating in time-frequency space. The method relies on an adaptive approximation of a signal by means of waveforms chosen from a very large and redundant dictionary of functions. The MP performance is illustrated by simulations and examples of sleep spindles and slow wave activity analysis. An improvement of the original procedure, relying on the introduction of stochastic dictionaries, is proposed. A comparison of the performance of dyadic and stochastic dictionaries is presented. MP with stochastic dictionaries is characterized by an unmatched resolution in time-frequency space; moreover it allows for parametric description of all (periodic and transient) signal features in the framework of the same formalism. Matching pursuit is especially suitable for analysis of non-stationary signals and is a unique tool for the investigation of dynamic changes of brain activity.

Phase and amplitude analysis in time-frequency space--application to voluntary finger movement.

Two methods operating in time-frequency space were applied to analysis of EEG activity accompanying voluntary finger movements. The first one, based on matching pursuit approach provided high-resolution distributions of power in time-frequency space. The phenomena of event related desynchronization (ERD) and synchronization (ERS) were investigated without the need of band-pass filtering. Time evolution of mu- and beta-components was observed in a detailed way. The second method was based on a multichannel autoregressive model (MVAR) adapted for investigation of short-time changes in EEG signal. The direction and spectral content of the EEG activity propagation was estimated by means of short-time directed transfer function (SDTF). The evidence of 'cross-talk' between different areas of motor and sensory cortex was found. The earlier known phenomena, connected with voluntary movements, were confirmed and a new evidence concerning focal ERD/surround ERS and beta activity post-movement synchronization was found.

Time-frequency microstructure of event-related electro-encephalogram desynchronisation and synchronisation.

A new method is presented for the analysis of event-related EEG phenomena, in particular event related desynchronisation (ERD) and event related synchronisation (ERS) related to a voluntary movement; the method offers: high time-frequency resolution and, hence, increased ERD/ERS sensitivity (especially in the gamma band, where improvement can exceed an order of magnitude); the ability to analyse the whole picture of energy changes at once, without setting a priori the analysed frequency bands; and a parametric description of the signal's structures. The main idea is based upon averaging energy distributions of single EEG trials in the time-frequency plane. As the estimator for the signal's energy density, matching pursuit is chosen, with stochastic Gabor dictionaries. Other possible estimates are presented on a simulated signal and discussed briefly. The consistency of the results with previous findings is evaluated on the data from a classical voluntary finger movement experiment.

High resolution study of sleep spindles.

OBJECTIVE: Universal high-resolution time-frequency parameterization of sleep EEG structures. METHODS: A new algorithm called Matching Pursuit was used for the decomposition of sleep EEG into waveforms chosen from a large and redundant set of functions. As a result all signal structures were parameterized in terms of their frequency, time occurrence, time span and energy. Slow wave activity and sleep spindles were identified according to neurophysiological criteria and various distributions describing their time evolution, topographical and frequency characteristics were constructed. RESULTS: Two types of sleep spindles of different topological and spectral properties were identified. High time-frequency resolution made possible separation of superimposed spindles. Cross-correlation between high- and low-frequency components of superimposed spindles revealed a fixed time-delay between them, the high-frequency component preceding the low-frequency one. CONCLUSION: The results of our study suggest that processes of generation of both types of sleep spindles are weakly coupled.

Time-frequency analysis using the matching pursuit algorithm applied to seizures originating from the mesial temporal lobe.

OBJECTIVES: The ability to analyze patterns of recorded seizure activity is important in the localization and classification of seizures. Ictal evolution is typically a dynamic process with signals composed of multiple frequencies; this can limit or complicate methods of analysis. The recently-developed matching pursuit algorithm permits continuous time-frequency analyses, making it particularly appealing for application to these signals. The studies here represent the initial applications of this method to intracranial ictal recordings. METHODS: Mesial temporal onset partial seizures were recorded from 9 patients. The data were analyzed by the matching pursuit algorithm were continuous digitized single channel recordings from the depth electrode contact nearest the region of seizure onset. Tine frequency energy distributions were plotted for each seizure and correlated with the intracranial EEG recordings. RESULTS: Periods of seizure initiation, transitional rhythmic bursting activity, organized rhythmic bursting activity and intermittent bursting activity were identified. During periods of organized rhythmic bursting activity, all mesial temporal onset seizures analyzed had a maximum predominant frequency of 5.3-8.4 Hz with a monotonic decline in frequency over a period of less than 60 s. The matching pursuit method allowed for time-frequency decomposition of entire seizures. CONCLUSIONS: The matching pursuit method is a valuable tool for time-frequency analyses of dynamic seizure activity. It is well suited for application to the non-stationary activity that typically characterizes seizure evolution. Time-frequency patterns of seizures originating from different brain regions can be compared using the matching pursuit method.

Time-frequency analysis of vibrotactile driving responses by matching pursuit.

A new method of time-frequency analysis, based on the Matching Pursuit (MP) algorithm, was used to extract and quantify EEG 'driving' or frequency-following responses produced in human primary somatosensory cortex (SI) by 33 Hz vibrotactile stimulation of the right index fingertip in a single subject. EEG signals were recorded from a 5 x 5 array of electrodes centered over the left hand area, time-locked to repeated presentations of four vibratory stimulus amplitudes. The MP algorithm was used to decompose the edited and and filtered EEG signals into waveforms selected from a large and redundant dictionary. Statistical discrimination of the vibratory stimulus amplitudes was then readily achieved in terms of trial-by-trial measures of response amplitude constructed in automated fashion from the calculated MP parameters. The results were orderly and physiologically coherent, and potentially open the way to correlation of psychophysical magnitude estimates with measures of neurophysiological response on a trial-by-trial basis. The approach developed here appears well suited to detection and characterisation of time dependent or transient target signals embedded in a noisy background.