Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 145

1966, Groves and Thompson, 1970]. Moreover, these authors proposed that response
to repeated stimulation is the interaction of sensitization and habituation.
In order to study these systematic changes over time, there are in general two
possible approaches. One is the adjustment of the experimental paradigm, and this
is described further in this paragraph. Second is the development of methods able to
extract single-trial responses; these are described in the following paragraph.
In the classical works on habituation/sensitization we can find two possible
paradigms: sub-ensemble averages and block averages. In case of sub-ensemble av-
erages recording sessions were subdivided into consecutive sub sets consisting of a
few consecutive trials. In each of the sub sets the ensemble average is computed.
The time evolution of these sub-ensemble averages is used to assess the habitua-
tion/sensitization process. This approach is successful only if changes between trials
are much slower than the time span of the number of trials included in each sub-
average. Another approach to study the trial-to-trial changes is to repeat a sequence
of trials as a block, each sequence starting after an assumed recovery time. Then, cor-
responding responses from each block are averaged. This second method, assumes
that consecutive blocks are equivalent and the habitation/sensitization processes are
exactly the same in each block. Moreover, both sub-ensemble and block-averaging
require long recording sessions, and thus general arousal changes are likely to occur
during the experiment [Quian Quiroga and van Luijtelaar, 2002].

4.1.7.2 In pursuit of single trial evoked responses

The averaging techniques do not allow for direct access to variations of ERP la-
tencies and amplitudes. In particular, short lasting changes, which may provide rel-
evant insights into cognitive operations, are blurred or even masked by averaging
processes, whereas single trial analysis would allow a reliable extraction of signal
characteristics from a single ERP sequence. From this point of view, single-trial
analysis of ERP is the ultimate goal of almost all related research, crucial for un-
derstanding subtle changes in individual brain responses to a repeated stimulus.
Several techniques have been proposed in order to improve access to individual
ERPs. In each case specific assumptions must be made about the nature of the re-
sponse.

4.1.7.2.1 Wiener filters Many approaches involve filtering of the single-trial
traces, in particular using techniques based on the Wiener formalism (Sect. 2.1.4),
which provides an optimal filtering in the mean square error sense [Woody, 1967,
Walter, 1969, Doyle, 1975]. However, these approaches have the common drawback
of considering the signal as a stationary process and, since the ERPs are composi-
tions of transient responses with different time and frequency signatures, they are not
likely to give optimal results.

4.1.7.2.2 Model based approach A model based approach was proposed in
[Cerutti et al., 1987]. The background EEG activity was modeled as an AR process
and event-related activity as autoregressive moving average process ARMA. In this