EEG and Brain Connectivity: A Tutorial - Bio-Medical Instruments, Inc.
EEG and Brain Connectivity: A Tutorial - Bio-Medical Instruments, Inc.
EEG and Brain Connectivity: A Tutorial - Bio-Medical Instruments, Inc.
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Eq. 1-<br />
r =<br />
∑<br />
N<br />
∑<br />
N<br />
( X<br />
( X<br />
−<br />
−<br />
X )( Y − Y )<br />
X )<br />
2<br />
∑<br />
N<br />
( Y − Y )<br />
2<br />
or the computationally simpler equation that one can compute more easily<br />
using a h<strong>and</strong> calculator:<br />
Eq. 2 -<br />
r<br />
=<br />
( N<br />
N∑ XY − ∑ X∑Y<br />
2<br />
2<br />
2<br />
∑ X − ( ∑ X ) )( N∑Y<br />
− ( ∑<br />
Y )<br />
2<br />
)<br />
For example, if one computes the FFT over 1 second epochs for a 60<br />
second recording period, i.e., N = 60, then the number of degrees of freedom<br />
in the spectral correlation coefficient (SCC) for channels X <strong>and</strong> Y = 60 – 1 =<br />
59. For 59 degrees of freedom then a correlation of 0.258 or higher is<br />
statistically significant at P < .05. This is a valid <strong>and</strong> commonly used<br />
connectivity measure, however, it is important to remember that the<br />
correlation coefficient includes volume conduction + network connectivity,<br />
i.e., they are inextricably confounded. This is because the correlation<br />
coefficient omits phase difference <strong>and</strong> involves the “in-phase” or<br />
autospectral values <strong>and</strong> therefore volume conduction can not be separated<br />
<strong>and</strong> eliminated. This makes it more difficult to know if factors such as the<br />
number <strong>and</strong> strength of connections are what are changing due to<br />
experimental control or is it the “volume conduction” that is changing? As<br />
explained in section 8, coherence using complex numbers <strong>and</strong> phase<br />
differences separate volume conduction from network dynamics <strong>and</strong><br />
automatically solve this problem.<br />
Another method of applying the Pearson Product correlation was<br />
developed by Lexicor, <strong>Inc</strong>. in the 1990s. This method computes the<br />
correlation between <strong>EEG</strong> spectra measured from two different locations <strong>and</strong><br />
uses the individual spectral bin values within a frequency b<strong>and</strong>. For<br />
example, if there are five frequency bins in the alpha frequency b<strong>and</strong> (i.e.,<br />
8Hz, 9Hz, 10Hz, 11Hz <strong>and</strong> 12Hz), then N = 5 <strong>and</strong> the number of degrees of<br />
freedom = N – 1 = 4. When the degrees of freedom = 4 then a correlation<br />
coefficient of 0.961 or higher is necessary in order to achieve statistical<br />
significance at P < .05. Equations 1 <strong>and</strong> 2 are used to calculate the Pearson<br />
product correlation in both instances.