Methods for monitoring effects of drugs and other chemicals in the CNS by using high resolution EEG

The estimation of brain connectivity allows describing the functional links established between different cortical areas during the execution of a particular experimental task and is an important step to the understanding of the brain functional organization. The importance of using non-invasive methods for the measurement of the brain activity has conveyed more and more attention on techniques like the Electroencephalography (EEG), the Magnetoencephalography (MEG) and the functional Resonance Imaging (fMRI). While the fMRI measurements allow for a very high spatial resolution (in the order of millimeters) but of a poor temporal resolution (order of seconds), EEG and MEG show a very high temporal resolution (in the order of milliseconds) with a poor spatial resolution (order of centimeters). To overcome the limitations of conventional EEG and MEG, a body of techniques was developed in the last 15 years to improve the spatial resolution of the EEG, under the name of High-resolution EEG [Babiloni et al, 2000]. These include the use of a large spatial sample of the EEG on the scalp, a multi-compartment head model (scalp, skull, dura mater, cortex) constructed from individual MRI from each subject, a distributed source model, and a regularized linear inverse source estimates of cortical current density (Linear Inverse Problem, LIP). The result is a reconstruction of the electrical activity at the cortical level with a spatial resolution that is greatly improved with respect to the conventional EEG. On the other side, there is an increasing interest for the use of mathematical algorithms aiming at estimating the flow of information between different scalp or cortical areas in humans [Baccalà, and Sameshima, 2001]. In fact, the methods proposed in literature typically involve the estimation of some covariance properties between the different time series measured from different spatial sites, during motor and cognitive tasks, by EEG and fMRI techniques. Due to evidence that important information in the EEG signals are coded in frequency rather than in time domain, attention was focused on detecting frequency-specific interactions in EEG or MEG signals, for instance by means of the coherence between the activity of pairs of channels [Kaminski and Blinowska, 1991). However, coherence analysis does not have a directional nature (i.e. it just examines whether a link exists between two neural structures, by describing instances when they are in synchronous activity) and it does not provide the direction of the information flow. In this respect, multivariate spectral techniques called Directed Transfer Function (DTF) or Partial Directed Coherence (PDC) were proposed to determine the directional influences between any given pair of channels in a multivariate data set. Both DTF and PDC can be demonstrated to rely on the key concept of Granger causality between time series, according to which an observed time series x(n) causes another series y(n) if the knowledge of x(n)’s past significantly improves prediction of y(n). This relation between time series is not reciprocal, i.e. x(n) may cause y(n) without y(n) necessarily causing x(n). This lack of reciprocity allows the evaluation of the direction of information flow between structures. These estimators are able to characterize both the direction and spectral properties of the brain signals, and require only one multivariate autoregressive (MVAR) model estimated from all the EEG channels. Several simulation studies clearly demonstrated that the estimation of cortical connectivity patterns could be performed with the DTF and PDC methods with a relative low amount of errors. The quality of these estimations depends on the particular length of the neurophysiologic recording under analysis, and on the level of the signal to noise ratio present in the dataset to be processed. In fact, under general conditions normally met in the experimental EEG or MEG recordings, the amount of errors performed by re-estimating the imposed connectivity pattern has been obtained below the 5% level [Astolfi et al., 2005]. Recent studies have stressed the limits of conventional pairwise methods with respect to the multivariate spectral measures based on the autoregressive modeling of multichannel EEG, in order to compute efficient connectivity estimates. Among the multivariate methods, the Partial Directed Coherence is an estimator characterizing, at the same time, direction and spectral properties of the interaction between brain signals, and requires only one MVAR model to be estimated from all the time series. However, the classical estimation of this method requires the stationarity of the signals; moreover, with the estimation of a unique MVAR model on an entire time interval, transient pathways of information transfer remains hidden. This limitation could bias the physiologic interpretation of the results obtained with the connectivity technique employed. To overcome this limitation, different algorithms for the estimation of MVAR with time dependent coefficients were recently developed. Moeller et al [Moeller, 2001] proposed an application to MVAR estimation of the extension of the recursive least squares (RLS) algorithm with a forgetting factor. This estimation procedure allows for the simultaneous fit of one mean MVAR model to a set of single trials, each one representing a measurement of the same task. In contrast to short-window techniques, the multi-trial RLS algorithm does not require the stationarity of the signals, and involves the information of the actual past of the signal, whose influence decreases exponentially with the time distance to the actual samples. The advantages of this estimation technique are an effective computation algorithm and a high adaptation capability. It was demonstrated in [Moeller, 2001] that the adaptation capability of the estimation (measured by its adaptation speed and variance) does not depend on the model dimension. Simulations on the efficacy of time-variant Granger causality based on AMVAR computed by RLS algorithm were also provided. The main message of this paper lies in the fact that with the body of technique knows as high resolution EEG and the PDC in both stationary and non stationary conditions, it is possible to describe the cortical and the functional connectivity pattern from EEG recordings during cognitive tasks. This assessment of cortical activity and connectivity is reliable, as supported by thousands of simulations performed and published. Such estimation could be essential to describe the brain activity in those pathologies that are aimed to be cured with drugs that moves inside the blood brain barrier. The Figure 1 illustrates the different steps that bring from the high resolution EEG recordings to the estimation of cortical connectivity while the Fig.2 describes the capability of the described methodology to estimate the functional cortical connectivity between brain areas. It is out of doubt that such kind of description of the EEG activity could be crucial in the evaluation of the effects of drugs on the CNS. In fact, it will be possible to describe how the cortical activity and connectivity changes during the therapy, before that these changes could produce appreciable changes in overt behavior of the patients