Analysis of EEG and MEG Data

In general, there are two major ways of interpreting EEG and MEG data; this dual approach can be understood to some extent by analogy to the wave and particle views in modern physics. One way is to consider EEG and MEG recordings as composed of discrete, relatively short-lasting neural events ("particles") triggered by specific events (plus "noise" not related to any event of interest). To enhance such neural events, the EEG or MEG can be time-locked to particular experimental events and averaged to yield ERPs or ERFs, respectively. Inferences from ERPs and ERFs are made primarily from their changes over time. In addition, the location of such changes often also provides useful data. The classical approach has been to define "components" in terms of their polarity [ERPs only: positive (P) and negative (N)], latency (milliseconds), scalp distribution, source location, and function (i.e., modulation with specific experimental manipulations). For example, the P150 (or ERF M180) is a broadly distributed peak that is positive (hence the P) at the vertex around 150 msec. This event has been found to index early perceptual categorization of well-learned visual stimuli (i.e., words and faces), which seems to reflect activity in specific parts of the brain (the posterior fusiform gyrus and occipito-temporal sulcal regions).

A common assumption is that each ERP or ERF peak reflects one or more perceptual or cognitive processes and, therefore, is a neurophysiological index of those processes. To determine whether two experimental conditions engage the same or different processes, researchers evaluate whether both conditions evoke the same ERP or ERF component, and the degree of activation can be inferred from peak amplitude or area. More important, the time course of activation of these neural processes can then be inferred from the peak latency or the time when the conditions diverge.

Another way to analyze EEG and MEG recordings is to consider them as long stretches of data produced by a large ensemble of neuronal oscillators ("waves"). This view has led to analyses of the power (amplitude) of different frequencies of oscillations, which shift over time and when people are in different mental states. The first major success of this approach was to reveal that sleep involves different stages, as indicated by shifts in the power of the oscillations at different frequencies.

Hybrid approaches integrate the particle and wave methods by using "wavelet filters" to extract frequency information from the EEG while retaining temporal information. These approaches allow the detection of frequency changes occurring immediately after an event and have been used to discover EEG correlates of specific cognitive processes (e.g., linguistic processes).

The statistical analysis of EEG and ERPs is not trivial because of the large quantity of data acquired in most studies: for a typical experiment samples are recorded at thousands of time points from 100 or more spatial locations. Most investigators compare conditions of interest by performing tests based on the general linear model on data from a subset or all recording sites within various time windows. Similar to other neuroimaging techniques, the probability of finding an effect just by chance (so-called type I error) increases with the number of comparisons. The probability of a type I error is influenced by several factors, including the spatial and temporal correla tions in the data. Although a few methods have been proposed to keep the overall probability of a type I error within acceptable limits, currently no universal solution exists.

Aspergers Answers Revealed

Aspergers Answers Revealed

Learn How to Help, Understand amp Cope with your Aspergers Child from a UK Chartered Educational Psychologist. Before beginning any practice relating to Aspergers it is highly recommended that you first obtain the consent and advice of a qualified health,education or social care professional.

Get My Free Ebook

Post a comment