Determination of cortical sources that significantly oscillated at
the stimulus driven frequencies (12 Hz and 15 Hz)
We determined which cortical sources showed oscillatory neural activity
at the driving ssVEF frequencies significantly different from background
noise. As power estimates are the sum of two squared measures (real and
imaginary parts), power estimates of the signal (here at the ssVEF
frequency of interest) and noise (here at frequency bins different from
the ssVEF frequencies of interest) follow a chi-square distribution
(Dobie and Wilson 1996). Therefore, the power at the frequency of
interest can be tested against noise by an F-ratio test with degrees of
freedom df = 2 for the numerator and df = 2m for
the denominator, whereby m is the number of noise frequency bins
(Dobie and Wilson 1996).
Using this method, we created two masks for the pre-cue central fixation
cross task at 12 Hz and 15 Hz (flicker frequency for each visual
hemifield was counter-balanced across subjects) based on the grand
average power spectrum across all subjects, representing our signal.
Noise was defined by 31 noise frequency bins between 17 Hz and 20 Hz
(0.1 Hz steps), that were chosen as they did not contain any harmonics
of the ssVEF responses. The sources included in the masks were
parametrically determined by an F test with the corresponding degrees of
freedom at a false discovery rate corrected alpha threshold of 0.05
(3003 comparisons, threshold at F(2, 62) = 9.03, pfdr
corrected < 0.001). Then, these two masks were OR logically
combined into one mask for the pre-cue central fixation cross task. The
same was done for the pre-cue peripheral rings task (threshold at F(2,
62) = 10.12, pfdr corrected < 0.001). The two
masks were created in order to compare the spatial extension of
significant ssVEF responses in cortical source space between these two
pre-cue baseline tasks (see Results).
Finally, a third mask was created by logically OR combining the source
masks of the F-ratio tests on the grand means of power spectra of the
fully balanced combinations of pre- and post-cue conditions (e.
g. pre-cue baseline attend rings, post-cue lvf not attended, post-cue
lvf attended, pre-cue baseline attend cross, post-cue lvf not attended,
post-cue lvf attended, etc.). This was done in order not to bias the
implicated sources in the mass-univariate statistical tests (see below)
towards one pre- and post-cue task combination. The same parameters as
described above were applied to these F-ratio tests (threshold at F(2,
62) = 8.69, pfdr corrected < 0.001). Figure 1
depicts the resulting overall mask for the overall statistical
comparisons (see below).