External Link

The ERP analysis will first focus on the modulation of the visual evoked potentials upon stimulus onset in the far configuration. At present, only data from Kurt has been looked at (Kurt 40-48 mt_k_006). This may expand to cue and attention detection effects in the future.

Every trial has an onset of stimuli. This could have been two stimuli (conditions 4-11) or one stimulus in the contralateral position (conditions 20,22,24,26) or one stimulus in the ipsilateral position (conditions 21,23,25,27). The data is cut from 0.5 seconds prior to the onset of stimuli, up until 0.8 seconds after stimuli onset.

The ERPs of these three different conditions were calculated. First a low pass filter was applied, passing below 35hz. Thereafter a visual artifact rejection at variance 5. Following this, manual artifact rejection of trials was conducted. One channel was rejected (A10). The bipolar data was calculated. After each of the above steps, the data was normalized. After artifact rejection, there were 2583 trials when two stimuli were present; 231 trials when only the contralateral stimulus was present; and 253 trials when only the ipsilateral stimulus was present.

Fig. 1: Comparison of multiple stimuli conditions and single stimulus conditions. Below is an example from channel J06-J05 in V1. The blue line shows the ERP when both stimuli are present. The red line shows the ERP when only the contralateral stimulus is present. The green line shows the ERP when only the ipsilateral stimulus is present. x-axis shows -0.05s prior to stimulus onset to 0.4s after.

|

With the ERPs calculated with both unipolar and bipolar data, it was first controlled that there was no difference in blue or yellow being in the contralateral position when there were two stimuli. This difference was negligible and the trails are henceforth analyzed together.

Fig. 1: Comparison of unipolar and bipolar ERPs in the two stimuli trials. (Left) Two unipolar ERPs (Red) from J06 and J05. The bipolar J06-J05 recording is shown in blue. The unipolar ERP from J06 minus the unipolar ERP from J05 is shown in black. The difference in amplitude between the black and blue is (most probably) due to a normalization that occurs after the bipolar is calculated. x-axis shows -0.05s prior to stimulus onset to 0.4s after. (Right) ERP shows unipolar G15, J14, G15 minus G14 and bipolar G15-G14. Line colors and x-axis are the same opposite figure.

Fig 2: Topographic comparison bipolar and unipolar at presentation of both stimuli. (Top-Left) Unipolar peak 1. Topographic image at the first ERP peak at unipolar J15 (0.064s) (Top-Middle) Unipolar peak 2. Topographic image at the second ERP peak at unipolar J15 (0.091s) (Top-Right) ERP of J15. (Bottom-Left) Bipolar peak 1. Topographic image at the first ERP peak at bipolar J15-J14 (0.067). (Bottom-Middle) Bipolar peak 2. Topographic image at the second ERP peak at bipolar J15-J14 (0.090). (Bottom-Right) ERP of J15-14.

Fig 1. and Fig 2. both suggest that bipolar ERPs are possible, at least for the early components.

The next stage of this analysis is to find the Granger of the fixation period occurring prior to the stimulus. The pre-stimulus period is from ca. 0.8s prior to stimulus presentation. There were some visual artifacts immediately after focus began. Because of this, The data was cut 0.6s prior to stimulus onset.

It was mentioned in a group meeting that it is generally better to use an odd number of tapers, where below only Hanning, 1, 2, 4, 6 and 8 multi-tapers were initially used. I have now run power with tapers 1-8 and the odd numbers slot in-between the even numbers perfectly. There does not seem to be a need for an odd number of tapers.

Fig 1: (Left) Comparison of the power for the different multi-tapers (Hanning, DPSS: 1-8) at H26-H25. (Middle) Comparison of multi-tapers (Hanning, DPSS: 1,2,4,6,8) coherence at H26-H25 with reference to J15-J14. (Right) Comparison of multi-taper (Hanning, DPSS: 1,2,4,6,8) strength at H26-H25.

Fig 2: Comparison of multi-tapers Granger at H26-H25 with reference to J15-J14. (Left) H26-H25 as the sink. (Right) H26-H25 as the source.

Fig 3: Comparison of multi-tapers summed granger at H26-H25. (Left) H26-H25 is the sink. (Right) H26-H25 is the source.

Interestingly, there is derfinitely some alpha in the Hanning taper Granger results. While other tapers (mt1 and mt2) have in in 2-left when H26-H25 is the sink, this is not the case in 2-right where H26-25 is the sender. Only Hanning picks up this alpha. This is probably a problem with the Hanning.

Using the preexistent function, a Fieldtrip-like wrapper has been made. This makes the single trial ERP analysis compatibility with configuration settings such as time-window (some more configuration settings should still be added, such as cfg.channel).

After artifact rejection on the Granger data there were 2322 trials of condition 4-11 remaining.

Fig 1: (Left) Topoplot showing the mean cross-correlation value of the single trial ERP fitting to the template ERP. (Right) Topoplot showing the amount of trials that had a cross-correlation greater than 0.8. .

| |

Here we order the amplitude scaling factor returned by the single trial ERP function. Only cross correlation values greater than 0.6 were included. From this, trials were places in bins of 100 trials. Each bin overlapped by 50 trials with the previous bin. The Granger for each bin was then calculated. This was the correlated with the average ERP amplitude scaling factor of the bin.

Fig 1 Non-zero correlations of all channels where p<0.000001 number of Granger correlations to binned ERP (Left) for all channels at each frequency; (middle) topographically showing the channels whose Granger “predicted” an ERP of a different channel at all frequencies. (Right) topographically showing the channels that had their ERP's “predicted” at all frequencies.

| | |

In the above figures, single channel Grangers did not correlate with the ERP amplitude in V1. We then calculated the overall sink Granger strength of each channel. This gives the overall input to that channel at a given frequency. When doing this, some V1 channels become correlated with the ERP amplitude at beta.

Fig 2 Non-zero Granger strength/single trial ERP amplitude correlations of all channels where p<0.01 between 13-21hz.

|

The woody filter calculates a scaling amplitude based upon the raw data's relation to the ERP. Wavelet denoising uses a function (here B-Splines) and gives a ERP-like wave reconstructed from the transformations of the B-spline function at different frequencies. In the above section, the Woody filter was used. Here we tested to see whether these two different methods resulted with the same analysis (i.e. are ethe same trials correlated.)

The wavelet single trials were calculated on 128 sample data between 0.034s after stimulus presentation and 0.161s after stimulus presentation. This time period was chose to make sure that P1 and N1 were always within the window.

There are several ways denoised wavelet single trials could be used for this comparision with the amplitude scaling factor of the Woody filter. At the moment I am using: Waveamp = abs(min) + abs(max). I.e. the amplitude scaling factor is the max value plus the absolute value of the minimum value.

Fig 1 Denoised single trials of channel J21-J20. Black line shows the ERP.

|

Fig 2 (left) p-value between Waveamp and amplitude scaling factor of the woody filter per channel (middle) Correlation between Waveamp and the amplitude scaling factor of the woody filter at J21-J20 for all trials. r=0.63 p=0. (right) Correlation between Waveamp and the amplitude scaling factor of the woody filter for trials where the cross correlation is greater than 0.6 at J21-J20. r=0.73, p=0.

| | |

K19-K18 is the only channel to have a p value above 0.05. K11-K10 has a p-value above 0.01.

Pre-Stim vs. Post-Stim

Gamma Power baselined to prestim period. Prestim period: -400-0. Poststim period: 200-400.

|

Clustering Coefficients of prestim vs. poststim calculated on the granger of the same period as above. The clustering coefficient of channel x is how likely two of channel x's “connections” also connect to each other. Reveals too different (Granger) networks active at beta before and after stimulus onset. (TOP) Prestimulus beta clustering coefficients. (Bottom) Poststimulus beta clustering coefficients.

|