Application to biomedical signals - Practical Biomedical Signal Analysis Using MATLAB - page 117

are mapped. The average power values in each frequency band and their ratios are
also useful markers in depression [Cook et al., 2009].
In some applications not only rhythmic activity, but also transient structures of
EEG are of importance. In this case the methods of non-stationary signal analy-
sis have to be used. The examples of such transients are sleep spindles or epileptic
spikes, which are important, e.g., for assessment of the influence of sleep-inducing
or anti-epileptic drugs. The optimal method for detection of transients occurring ran-
domly in signals is matching pursuit. The advantages of the method in estimation
of influence of sleep-inducing drugs was demonstrated in [Durka et al., 2002], for
identification of spindles in [Zygierewicz et al., 1999], and for detection of epileptic
spikes in [Durka, 2004].

4.1.6.2 Multiple channel analysis

4.1.6.2.1 Mapping One of the commonly used methods of presenting multichan-
nel brain activity is the topographical display of signals features called mapping. A
map may help to make a direct comparison between the topographic distribution of
EEG features and an anatomic image given, e.g., by the tomographic brain scan.
Three types of features are most commonly mapped for clinical applications: 1) di-
rect variable such as amplitude, 2) transformed variable such as total spectral power
or relative spectral power in frequency band, 3) the result of statistical test applied
to a given EEG feature. The amplitude values of the signals or the spectral power in
selected bands are frequently displayed. To obtain a map composed of hundreds of
pixels spatial interpolation between the discrete electrode positions is necessary.
Different interpolation algorithms are possible, ranging from the simple N-nearest
neighbors electrode algorithm to the more complex spline approximation. The ad-
vantage of spline interpolation methods is the easy computation of the second spatial
derivative and thus application of the Laplacian operator. In MATLAB the interpo-
lation of maps can be done, e.g., with the interp2 function, which supports nearest
neighbor, linear, and cubic spline interpolation. For more sophisticated spline inter-
polations the Spline Toolbox may be useful.
One should not assume that maximum of signal power visible on the map corre-
sponds to the underlying source, since it can come from superposition of activity of
coherent sources. The character of the map is strongly dependent on the reference.
As was already mentioned, there is no optimal reference system. In case of mapping
one has to be aware of this limitation. Common average reference system is liable to
serious ambiguities in cases when some of the electrodes pick up similar signals, e.g.,
occipital alpha rhythm may appear in frontal derivations and eye movement artifacts
may affect posterior electrodes [Pfurtscheller, 1991].
The recommended representation involves surface Laplacians ( Sect. 4.1.3) . How-
ever, a reliable computation of surface Laplacian requires at least 64 electrodes and
adequate spatial sampling is obtained for more than 128 electrodes. Therefore, quite
frequently an approximation of the Laplacian operator by Hjorth transform is ap-
plied. Results obtained by application of the Laplacian operator may be further ame-
liorated by deblurring; that is, using a mathematical model of volume conduction