In Fig. 1. we observed that SNR of CorrCA components is significantly higher than channel's SNR in both target and non-target trials. Since EEG channels always suffer from a low signal-to-noise ratio, the significant CorrCA component can be used as a feature representing the P300 component. Poor SNR might indicate the presence of noise in both channels and components. We observed that after the third CorrCA component, other components became nonsignificant. So, we rejected those components for further analysis. The SNR of the components is also higher than the SNR of the channels.
In Fig. 4 the first component showed a sharp peak around 400 ms and captured the P300 component. We observed that only the first CorrCA component represented the P300 peak latency. So, we can use only the first component instead of eight channels. Figure 1. showed only the first three components are significant from the target trials. This removes the ambiguity that arises in the case of channel selection for machine learning applications.
Previous work considered CorrCA components corresponding to three highest ISC values [32]. In this work for each subject, we took sum of ITC scores of first three CorrCA components. We computed the sum for both target trials and non-target trials cohort separately. ITCTarget (mean = 0.17, sd = 0.07) denotes the magnitude of correlated EEG activity with subjects across target trials cohort. ITCNontarget (mean = 0.064, sd = 0.03) represents the magnitude of the correlated EEG activity with subjects across non-target trials cohort. We performed a statistical t-test across total ITC scores of first three CorrCA components between target and non-target cohorts for all 8 subjects. It was observed that ITC Target values are significantly higher than the ITCNontarget (t(7) = 6.45, p < 0.01] and had a large effect size (Cohens' d = 1.39). It showed that the subjects were more attentive at the target stimulus than in the non-target stimulus. We also investigated whether ITC of individual CorrCA components could be influenced by attention. ISC and ITC are calculated in a similar way, except for the fact that ISC is computed across subject EEG data [33]. Computed ITC of Target trials cohort and Non-target trials cohort were significantly different from each other. We observed for first CorrCA component (t(7) = 5.91,p < 0.01,Cohens’ d = 1.25), second(t(7) = 6.96,p < 0.01,Cohens’ d = 2.01) and on third component (t(7) = 3.71,p < 0.01,Cohens’ d = 1.48). It was reported in[12] that subject 6 attended to non-target stimulus during one of the sessions, and subject 9 suffered from fatigue during the experiment, which might reflect their respective lower ITC. Subject 8 showed a higher ITC value during the entire session among all the subjects. It was reported that subject 8 was highly concentrated and motivated during the experiment[12]. All the disabled subjects showed lower averaged ITC values than healthy subjects.
We illustrated topoplots considering 8 channels, mainly from the parietal and central regions. Figure 5. and Fig. 6. shows the spatial patterns of neural activity in target and non-target cohorts using forward model. Here, A1 represents the neural correlation with the first CorrCA component, and similarly A2 represents neural correlation with second CorrCA component. Each topoplot has 8 weights corresponding to the 8 channels. In both Figs. 1 = P7, 2 = P3, 3 = Pz, 4 = Oz, 5 = P4, 6 = P8,7 = Fz, 8 = Cz. In Fig. 5 left panel showed neural activities after averaging target trial cohort for all subjects. Cz electrode captured the highest neural activity for able-bodied subjects (Row 1, right panel). From Fig. 5 we observed that higher neural activation was captured by first component for able, disabled and all subjects (Row 1). We observed that forward model of first component from target trials showed higher neural activation in able-bodied subjects (Row 1, right panel) than disabled subjects (Row 1, middle panel). We also observed that each component was able to capture a different pattern of neural activities. Figure 6 illustrates the spatial distribution of neural activity during the presentation of non-target images.