Peculiarities of information transfer within functional cortical network during emotional face perception

Aim: The research aimed to study and model the emotion-related activity of functional networks within the human brain cortex using power spectrum density and detrended phase transfer entropy methods. Attention was focused on revealing alterations in cognitive mechanisms, caused by presentation of neutral human faces as rare stimuli among faces with either negative or positive expression. Methods: EEG-data was recorded during the perception and processing of neutral human facial expressions, presented among positive and negative faces in two series of images, alongside with resting state with open and closed eyes, which was further analyzed using power spectrum density and detrended phase transfer entropy methods. Results: Specific EEG-bands (θ and β) were chosen for the analysis based on their prominent role in memory- and emotion-related mechanisms. The topography of the spectral power density corresponded to the generally accepted ideas describing perception and visual stimuli processing mechanisms. The phase transfer entropy method was not sufficient to analyze resting state data. The results of the analysis performed using the phase transfer entropy method revealed the problems of neutral faces differentiation when presented in a positive emotional context. Simultaneously, enhanced processes of motivational coding and self-reflection were observed during the presentation of neutral faces in a negative emotional context. These results corresponded with the data obtained in our previous ERP-based study. Conclusions: Phase transfer entropy and spectral power density have demonstrated their effectiveness in analyzing the mechanisms of emotional visual stimuli processing mediated in different cortical areas.


Introduction.
Human facial expressions are complex multi-dimensional visual stimuli, which provide the brain with a wide range of characteristics to process. The current study focused on the activity in and β bands of EEG data, as θ-and β-band oscillations directly reflect such cognitive processes as retrieval and actualization of memory [1], emotional excitement and other consciousness-driven processes [2]. Thus, changes in power spectrum density in these bands were measured, and effective connectivity was modeled using phase transfer entropy (phase TE).
There is a hypothesis that the brain has an internal model of the external world, which functions under Bayesian system principles while processing sensory input so that experience-modified response is elicited [3]. In our previously conducted ERP study, we have revealed that the stimuli' valence, which creates the emotional context for the target neutral expressions, affects perception. This effect is signified in the increase of attention level and memory processes [4]. To reveal and quantify "causal" or directional inter-areal phase-phase interactions during emotion perception, a new information theory-based approach of phase transfer entropy (phase TE) was used, as phase synchronization of neuronal-based oscillations has been suggested to determine the coordination and integration of anatomically distributed information processing [5].
Phase synchronization and amplitude correlations are functionally independent phenomena [6] and reveal different neural network functioning aspects. Besides, the importance of phase-derived information in neuronal processes is highlighted by studies showing that phase-based grouping can encode more information than the amplitude-based one in both visual and auditory [7] processing. This marks the oscillatory phase as a reflection of neuronal synchronization and a robust information transmitter between neural ensembles. The phase-derived information flow from one cortical region to another cannot be estimated using phase synchrony metrics [8], which, by their nature, are not able to reveal spatial orientation.
Phase transfer entropy (phase TE) [9] is a reformulation of the Wiener principle within the framework of information theory (IT) [10]. Like Granger's causality, TE assesses whether the past of the original and target time series affects predicting the target time series's future.
Contrariwise, phase TE compares conditional probabilities using the Kulbak-Leibler divergence. If signal X triggers the signal Y, then the probability density of the future Y, due to its past, must be different from the probability density of the future Y due to the past of both X and Y. Besides, unlike Granger's causality (and dynamic causal modeling), the phase TE is model-free as it does not carry any assumptions about the signal structure.
Phase TE can be a good indicator for analyzing phasebased connections and detecting directional interactions in broadband MEG / EEG sensor and source signals. However, the calculation of phase TE for individual tests requires a large amount of continuous data, which can be problematic in the case of a temporary task-related connection. The method can also be used for continuous data analysis with the help of several techniques of state space reconstruction.
This research aimed to study and model the effect of positive and negative faces on the perception and processing of target neutral faces. To do so, we focused our attention on determining the spectral power density and establishing causal relationships within θ-and β-bands, which play a unique role in human cognitive activity. Theta oscillations generated in the limbic system are considered the "emotional" band of the human brain. Thus, this activity reflects the cognitive component of the emotional reaction [11]. Besides, according to modern ideas, the increase of θ-activity in the anterior cortical areas can be assessed as a marker of enhanced activation, accompanied by inner attention and positive emotional experience [12]. Slow θ-oscillations are associated with memory-related processes and awareness, while high-frequency θ-oscillations reflect cognitive activity's emotional background. At the same time, β-band today is mainly associated with various aspects of brain function, from simple sensory responses (visual, auditory, somatosensory, etc.) to higher cognitive functions such as sensory memory, mechanisms of visual attention regulation, movement, emotional states, and execution of cognitive or creative tasks [13].

Materials and methods. Forty students of Taras
Shevchenko National University of Kyiv (21 females, aged 18-24, mean=21) were presented with two series of images, during which EEG was recorded. Data was also obtained during resting state with both closed and open eyes. The image demonstration procedure and recording of the cerebral cortex's induced activity were performed using the software and hardware complex "Neurocom" (KhAI Medica, Kharkiv, Ukraine) according to a specially created template. Electrodes were applied to the scalp following the international "10-20%" system.
Stimuli were selected from the International Affective Pictures System (IAPS) [14]. Emotional stimuli were shown in an arbitrary pattern in which the likelihood of a rare stimulus (n=100) appearance was 30%.
Transfer entropy (TE) is an information-theoretical measure that follows from the theory of information exchange to estimate conditional transition probabilities between two paired processes that develop over time.
The dPTE coefficient for the two cortical areas was calculated as the sum of the values of the PTE coefficients in both directions In order to compare the phase transfer entropy method to the existing methods of a quantitative assessment of human brain activity during cognitive activation, the spectral power density was calculated based on the fast Fourier transformation. This characteristic is the de facto standard measure of the strength of oscillation activity, which has been widely used in the study of cortical activity. Each EEG channel's spectral power density in each frequency band during each trial was calculated using the Welch period method [15]. The obtained values in the corresponding frequency range were normalized relative to the maximum value for emotional load and background, after which a map of the topographic distribution of the corresponding values was plotted on the scalp surface.
Results and discussion. The nature of the distribution of power spectrum density during resting state with both closed and open eyes can be described according to the general framework of the human brain's functioning in the absence of an urgent external task, brought out by the action of so-called default neural networks. Namely: increased power in θ1 and θ2 subbands in frontal (55.7 μV 2 / Hz and 54.6 μV 2 / Hz, respectively), occipital (55.3 μV 2 / Hz with eyes closed and 53.1 μV 2 / Hz with eyes open in θ1; 54.8 μV 2 / Hz with eyes closed and 53.4 μV 2 / Hz with eyes open in θ2 eyes) and parieto-central areas (maximum values up to 55.8 μV 2 / Hz in θ1 and 54.8 μV 2 / Hz in θ2) reflected the processes associated with memory, emotional manifestations, passive scanning of information (Fig. 1).
At the same time, the resting-state oscillations in β-band were characterized by a somewhat different topographic distribution: increased power of β1-oscillations was concentrated in the occipital regions of the cortex during the trial with closed eyes. However, eyes opening caused the spread of generalized activation around the cortex, thus covering not only occipital (up to 48.6 μV 2 / Hz) but also parietal, central, and frontal areas of the cortex (maximum values registered for frontal areas -48.4 μV 2 / Hz), with right-hemisphere lateralization in temporal regions (up to 48.5 μV 2 / Hz in the temporal area of the right hemisphere). As for the β2 frequency subband, the distribution of the increased oscillatory power in the occipital (up to 44.3 μV 2 / Hz), parietal (up to 43 μV 2 / Hz), and frontal (up to 44.3 μV 2 / Hz) areas of the cortex during resting state with closed eyes narrowed to the frontal regions (up to 44.1 μV 2 / Hz) after opening the eyes (Fig. 1).
Therefore, the resulting picture can be explained using existing literature data on these subbands' properties: traditionally, β1-subband is associated with external attention, cognition per se, while β2-oscillations are considered to reflect internal awareness, maintaining the status quo of brain activity. That is why topographic maps obtained in different resting-state trials can be juxtaposed with each other.
It is also worth noting that the obtained topographic maps of activation-related changes in power generally corresponded to the widely accepted visual stimuli processing framework. Namely: high-power foci in the prefrontal and fronto-central regions of the cortex in θ1 (up to 61 μV 2 / Hz), θ2 (up to 62.7 μV 2 / Hz) and β2 (up to 50.7 μV 2 / Hz) subbands, were accompanied with high values of oscillation power in the frontal, central, parietal and right-hemisphere occipital regions (values ranged from 48.2 μV 2 / Hz to 60 μV 2 / Hz) in β1-subband (Fig.1).
Concerning the results obtained by the detrended phase transfer entropy analysis in the processing of resting-state records with both closed and open eyes, it should be noted that the values of dPTE in such conditions were close to zero. This fact once again confirms the hypothesis of this method's effectiveness for analyzing only the states associated with cognitive activity. The connections formed within functional neural networks are shown in Fig.2 (dPTE≥0.2). The topography of PSD activation changes corresponded with prevailing views on visual stimuli perception and processing mechanisms. In both θ1-and θ2-subbands, extensive networks of effective connections during the processing of neutral faces in a positive emotional context were formed. However, in θ1-subband, it was generalized when in θ2-subband prominent activation nodes were formed in frontal, central, and parietal cortical areas in both hemispheres. As for neutral stimuli in a negative context, well-established nodes of activation in the left parietal and the cortex's central regions in θ2-subband were observed (Fig.2).
No stable network was observed for high-frequency oscillations in both trials, which corresponds with the concept of it as a marker of inward-directed attention. Simultaneously, in β1-subband, a more comprehensive network of causal connections with loci in parietal, frontal, and central regions was observed for target stimuli in positive context perception compared to the negative context.
Results obtained with phase TE during resting states with both closed and open eyes were generally close to zero, which supports the view that this analysis method is suitable for mapping brain networks associated with a specific cognitive task. However, utterly different connectivity patterns (C≥ 0.2, p≤ 0.05) were observed during stimuli demonstration, and they correlated well with the results of our ERP study [4].
Activation loci formed in θ2-subband (Fig.2) reflected memory retrieval processes, implicit encoding, differentiation, and integration of emotionally salient sensory information for positive context [16] and reflective mind notion and motivational significance encoding for negative context [17]. The endogenous self-referential processing of the CMS (cortical midline) has been related to an intrinsic virtual models generation, through which the brain forms an inferential knowledge about the structure of the environment. Moreover, well-pronounced connections among parietal regions of the cortex may manifest ongoing estimation of the person's social status depicted [18], which correlates with ERP late posterior negativity (1000 ms) in parietal areas, manifesting semantic component of stimuli perception. Differentiation problems (interference) and attention modulation were displayed through a vast network of connections in the β1-subband when neutral stimuli were presented in a positive context [19].

Conclusions.
In sum, phase TE and PSD demonstrated their effectiveness in analyzing emotional visual stimuli processing mechanisms mediated by disparate cortical areas. Consequently, PSD distribution corresponded with conventional views on visual stimuli perception and processing mechanisms. Withal, phase TE demonstrated no efficiency in resting-state data analysis. Phase TE revealed difficulties in neutral faces differentiation in a positive emotional context, alongside. Moreover, increased selfreflection and motivational encoding were observed when neutral faces were presented in a negative emotional context.