Figure 1.
(A) One of four possible stimuli (see middle) is presented visually for 200 ms. Subjects had to report the identity of the attended letter using a button-press. After a variable inter-trial-interval the next stimulus was presented. The stimuli were large (“global”) letters made up out of smaller letters (“local”). If the global and local levels spell different letters this creates a conflict and a decrease in reaction time. (B) Fluctuations in reaction time across the duration of the experiment were used as indicators of changes vigilance (shown here for one run of one subject).
Figure 2.
Optode positions (16 sources; red and 16 detectors; blue) and 44 measurement channels (green) on the surface of a 3D brain (left: view on parietal, right: on frontal regions).
Based on previous studies [3], [5] the optode positions were focused on parietal and prefrontal cortical regions.
Table 1.
The MATLAB toolbox NFRI [43] was used to estimate the MNI coordinates of the used EEG 10–20 positions.
Figure 3.
Timelines of the averaged accuracies (averaged Fisher-Z normalized correlation) of the prediction of subjects' single trial reaction times.
Different low-pass cutoff frequencies were chosen during the preprocessing (0.1 Hz, 0.15 Hz, 0.2 Hz, 0.25 Hz, and 0.3 Hz). Decoding performance on low-pass filtered data with cutoff frequencies between 0.15 Hz and 0.3 Hz is very similar. When the data are low-pass filtered with a cutoff frequency of 0.1 Hz decoding performance declines.
Figure 4.
Performance (left) and reaction times (right) for the congruent (blue) and incongruent (red) conditions. Responses to the congruent stimuli had a significant higher hit rate and were given faster compared to the incongruent stimuli. Reaction times to both the congruent and the incongruent stimuli were slower in the second part of the experiment (bottom). Between the third and the fourth run the instruction was to shift the attention from global to local or vice versa and potentially caused the slow response time in run 4. Error bars represent standard error of the mean.
Figure 5.
For illustration purposes the FIR-parameters for an HbO-channel in parietal (left) and prefrontal (right) areas are plotted. A) Time courses of FIR-parameter estimates for 2 reliable (based on a t-test on the parameter estimates) HbO-channels in parietal (left) and prefrontal (right) areas (the selected channels are marked with black circles in B below). Red: FIR parameter estimates of the ER-HR; Green: FIR parameter estimates of the ER-PM. The ER-HR can be thought of the mean averaged response to the stimuli and the ER-PM is the reaction time dependent modulation of the amplitude. For the two channels this means that for long reaction times the amplitude of early time points will be reduced and for late time points the amplitude will be increased. Please note the difference in amplitude between the prefrontal and parietal channel. B) FIR results of event-related hemodynamic response (ER-HR, top) and event-related parametric modulation (ER-PM, bottom) for the most significant time points (6 and 7 s after stimulus onset for the ER-HR and ER-PM, respectively) plotted on a 3D head surface. Black circles indicate the selected channels for which the time course is plotted in A.
Figure 6.
Timeline of the averaged accuracy (averaged Fisher-Z normalized correlation) of the prediction of subjects' single trial reaction times. During the time points that are marked with red asterisks (the time window that is highlighted with the red bar) decoding was significant (p<0.05; t-test on the Fisher-Z normalized correlation) above chance level.
Figure 7.
Average SVR weight vector for the most significant time point plotted on the surface of a 3D brain. The weights constitute a filter where channels with high or low values have a strong influence on the support vector regression. We found channels contributing to the prediction in both HbR (bottom) and HbO (top), and in prefrontal (left) as well as in parietal (right) brain regions.
Figure 8.
Result of the time-frequency decoding.
Averaged decoding performance (averaged Fisher-Z normalized correlation) for subjects' single trial performance for individual frequencies (wavelet transformation). Most information was encoded 5 s after stimulus onset in the slow frequency range between 0.1–0.15 Hz.