Time-frequency analysis

Each subject’s continuous data files were first downsampled from 1000 Hz to 250 Hz. This reduced the size of data under the consideration of the original bandwidth of signal. EEGLab tool (http://sccn.ucsd.edu/eeglab/index.html) was then used for filtering(Delorme & Makeig, 2004). The filter setting was a band-pass filter with

passband of 1-80 Hz. A notch filter was also set at 60 Hz to reject the interference of noise.

Thereafter we epoched data from -100 msec before stimulus onset to 500 msec after stimulus and set -100 msec to 0 msec as baseline. Removed baseline was performed in each epoch. Time-frequency analysis was performed using ERPwavelab

(http://www.erpwavelab.org/) (Morup et al., 2007). Since MMN is typically measured at the Fz electrode, only the Fz channel was sent in further calculation (Niznikiewicz et al., 2004; Javitt et al., 2008; Light et al., 2010).

Complex Morlet wavelet were used to compute the time-frequency analysis for each epoch and explored the six parameters(Morup et al., 2007). These six parameters from ERPwavelab included: ERSP, WTav, avWT, induced activity, ITPC and ITLC, could be explained as follows:

ERSP (event related spectral perturbation): The measure of the average power over epochs. The calculation of ERSP is shown in the following:

∑

where X(c, f, t, n) denote the time-frequency coefficient at channel c, frequency f, time t and epoch n of the signal given by x(c, t, n). The ERSP represents event-related (which is dependent on the different auditory stimulation type) spectrum perturbation, it represents the spectrum of this particular event-related potential (ERP) which deviant from the baseline brain activity. As its calculation method from the formula, it related to the

“distance” deviate from the baseline.

WTav: The measure of power in the average amplitude of the epochs which is denoted as

The WTav represents mean waveform from all trials for this particular ERP, which may cancel out the out-phase activity in each trial, then take the wavelet transformation and display its magnitude of spectrum. The physiological meaning represents the spectrum of the in phase brain activity for this particular ERP in whole trials.

avWT: The mean power from each time-frequency transformed evoke potential in each epoch which is denoted as follows:

⁼

∑

The avWT represents take all trials’ wavelet transform firstly then calculate their mean activity. The physiological meaning represents the mean sum spectrum of both in-phase and out-phases brain activity in whole trials.

Induced activity: The measure of non-phase locked activity. The induced activity represents difference between WTav and avWT, as shown in the following:

) This parameter could be used as the parameter of out-phase brain activity in ERP.

ITPC (inter-trial phase coherence): The measure of phase consistency over epochs which is denoted as follows:

∑

The ITPC is phase-lock index measure the phase consistency of whole epoch. Its

physiological meaning is phase consistency in difference frequency during whole epoch.

ITLC (inter trial linear coherence): The measure of phase consistency over epochs but weights epoch according to amplitude, as shown in the following:

∑ ∑

The ITLC is phase-consistency which weighted whole epoch’s amplitude. Its physiological meaning is weighted phase-consistency of this ERP.

5.2.5 Statistical analysis

We used the Statistical Package for the Social Sciences (SPSS) to carry out the comparison between 2 groups in ERP averaging and Time-frequency analysis, while the alpha level of 0.0025 was used.

A specific time-frequency region was further analyzed from the post-stimulation 100 ms to 350 ms and the frequency band between 1 Hz to 5 Hz which include the most MMN waveform power. For each individual, mean value within this time-frequency region in above 6 parameters was calculated for standard and oddball stimulation.

Statistic analysis was performed to compare the group difference in each parameter. To further study the relationship between the WTav, the induced activity and inter-trial phase coherence measurements in both controls and schizophrenia patients, we use the linear regression model to set WTav as a dependent variable and group effect, induced activity power and ITPC measurements as the independent variable.

5.3 Results

Demographic data of our subjects and results of traditional (grand average) ERP approach were shown in the upper part of Table 5.1. The controls showed significant younger but higher education level than the schizophrenia patients. Traditional ERP approach discovered group difference occurred in MMN mean amplitude in Fz electrode, which was compatible with previous literature.

Regarding the time-frequency results (Figure 5.1 and lower part of Table 5.1), the (two left) time-frequency plots for the standard stimuli were similar between control and schizophrenia subjects except for the avWT, which indicated the mean power from each time-frequency transformed evoke potential in each epoch. There was no significant difference in ERSP, WTav, induced activity, ITPC and ITLC of the standard stimuli between control and schizophrenia subjects.

In contrast, comparing (two right) time-frequency plots for the deviant stimuli between control and schizophrenia subjects, there is a decreased avWT (gray arrow) power in lower frequency range especially in the post-stimulation 100-350 msec in schizophrenia patients. Although there is no significant, the associated decreased coherence demonstrated by lower ITLC (blue arrow) and decreased out-phase activity power (avWT) provide the clue that schizophrenia subjects showed decreased oddball response could be explain by the desynchronized and decreased response of evoked potential. To further elaborate this issue, a regression model was use to study the relationship between the group effect, the parameters of phase-locked activity (WTav), the non-phase locked activity and inter-trial coherence (ITPC).

Using linear regression model to study the WTav in controls and schizophrenias, we found that after control the group effect, there is a significant negatively correlation with induced activity (p < 0.0001) and positively correlation with the ITPC (p < 0.0001) (Figure 5.2). The whole model p-value in both regression lines is below 0.0001 and R square is equal to 0.48 and 0.38 respectively.

5.4 Discussion

Two different approaches (averaged ERP, and time frequency analysis) were used to analyze MMN in schizophrenia patients and controls. Traditional ERP approach

discovered group difference occurred in MMN mean amplitude, which was compatible with previous literature (Michie et al., 2002; Niznikiewicz et al., 2004; de Wilde et al., 2007; Turetsky et al., 2007; Javitt et al., 2008; Keshavan et al., 2008).

The traditional signal averaging approach treats the oscillatory EEG activity as background “noise” in which the ERP “signal” is embedded, and discards the essential information(Delorme & Makeig, 2004). This missing information could be discovered using time-frequency analysis, which exhibits the underlying brain functions and their disturbances in schizophrenia patients. In our present study, we demonstrate that the group difference found in traditional approach can be further confirmed by

time-frequency analysis. We also found that using the regression model approach, the measure of power in the MMN (e.g. WTav) in both group can further explain by the induced activity and coherence measurement, e.g. the higher the WTav, the lower the induced activity and higher the coherence measurement. These results could further explain that in schizophrenia subjects have lower ERP in oddball stimulation may relate to the higher induced activity and lower inter-trial phase coherence than controls.

In our current study, the demographic data showed significant age and education difference in patients and controls, which may confound our findings in current work.

Besides, only the Fz channel was used in this time-frequency analysis, which may not clearly demonstrate the topographical change in the different brain regions. The whole picture of the brain activity difference in MMN could not be known. Further work should be address in the data collection and more comprehensive time-frequency analysis in whole EEG channels.

In summary, time-frequency approach explores the basic integrated neural network activity. It may also contribute to a better understanding of schizophrenia's essential pathology and the neurophysiological underpinnings in information processing.

5.5 Tables and Figures

Table 5.1 Demographic Data, ERP and Time-Frequency Results in Control and Schizophrenia Subjects.

Figure 5.1 Time-frequency results of MMN. The two upper graphs show the time-frequency plots in controls, for the standard (upper left panel) and the oddball (upper right panel) stimuli. The two lower graphs reveal the time-frequency plots in schizophrenia subjects, for the standard (lower left panel) and the oddball (lower right panel) stimuli.

Figure 5.2 Scatter plot of regression model between WTav, induced activity and ITPC within controls and schizophrenia subjects. The upper graph shows the WTav significantly negative correlation with induced activity. The lower graph reveals WTav’s significantly positive correlation with ITPC.