Akademik Veri Yönetim Sistemi
Cognitive Neurodynamics, cilt.18, sa.1, ss.49-66, 2024 (SCI-Expanded, Scopus)
https://link.springer.com/article/10.1007/s11571-023-09931-5
ABSTRACT
© 2023, The Author(s), under exclusive licence to
Springer Nature B.V.
The present study tests
the hypothesis that emotions of fear and anger are associated with distinct
psychophysiological and neural circuitry according to discrete emotion model
due to contrasting neurotransmitter activities, despite being included in the same
affective group in many studies due to similar arousal-valance scores of them
in emotion models. EEG data is downloaded from OpenNeuro platform with access
number of ds002721. Brain connectivity estimations are obtained by using both
functional and effective connectivity estimators in analysis of short (2 sec)
and long (6 sec) EEG segments across the cortex. In tests, discrete emotions
and resting-states are identified by frequency band specific brain network
measures and then contrasting emotional states are deep classified with 5-fold
cross-validated Long Short Term Memory Networks. Logistic regression modeling
has also been examined to provide robust performance criteria. Commonly, the
best results are obtained by using Partial Directed Coherence in Gamma
(31.5-60.5 Hz) sub-bands of short EEG segments. In particular, Fear and Anger
have been classified with accuracy of 91.79%. Thus, our hypothesis is supported
by overall results. In conclusion, Anger is found to be characterized by
increased transitivity and decreased local efficiency in addition to lower
modularity in Gamma-band in comparison to fear. Local efficiency refers
functional brain segregation originated from the ability of the brain to
exchange information locally. Transitivity refer the overall probability for
the brain having adjacent neural populations interconnected, thus revealing the
existence of tightly connected cortical regions. Modularity quantifies how well
the brain can be partitioned into functional cortical regions. In conclusion,
PDC is proposed to graph theoretical analysis of short EEG epochs in presenting
robust emotional indicators sensitive to perception of affective sounds.
CONCLUSION
From methodological point of view, the most meaningful
results are obtained by using PDC where PC can also provide useful results in
discriminating emotional states induced by fast-paced and emotionally
stimulating acoustic sounds. Unlike DTF, PDC is normalized to show a ratio
between the outflow from channel j to channel i to all the outflows from the
source channel j, so it emphasizes rather the sinks, not the sources. Therefore
PDC is found to be superior to DTF in the present study. In detail, PC measures
statistical association between two EEG segments based on covariance showing
the degree to which the amplitude change in an EEG segment is relative to the
amplitude change in another EEG segment, while PDC measures the normalized
relative coupling strength of frequency domain interactions from the source of
an EEG segment to the others across the cortex. Besides, the common technical
outcome of PC and PDC is to provide normalized results lie between 0 and 1. In
depth, PDC can quantify immediate directional coupling between neural
populations whereas DTF describes the existence of directional signal
propagation even if neural info travels through intermediate pathways rather
than through an immediate direct causal influence path. From computational
point of view, DTF estimates causal influence of an EEG recording channel on
the other one at particular frequency, while PDC provides only direct flows
between them. Regarding DTF, the resulting connectivity levels lie between 0
and 1 producing a ratio between the inflow (from a channel to a particular
channel) to all the inflows to a particular channel. In contrast to DTF, the
resulting estimations show a ratio between the outflow from a source channel to
the particular channel to all the outflows from the source channel in examining
PDC. Therefore, PDC provided better results for emotion recognition from EEG
recordings based on GT, while DTF was found to be useful for detection of
marginal variations such as local seizure (Franaszczuk and Bergey 1994), sleep stages (Kaminski and
Blinowska 1997) in past.
Several neuropsychiatric disorders cause both
functional and structural network parameters of segregation and integration to
change (Mu 2018). Dysfunctional brain connectivity can be
modeled by a loss of small-world organization of the brain, if healthy
connectome is correlated with balanced neural communication across the cortex.
By means of EEG recordings, graph theoretic functional brain parameters are
originated from superposition of excitatory and inhibitory post-synaptic
potentials that are caused by neurotransmitter release. The healthy brain
integrates a wide range of incoming external stimuli through binding the
spontaneous complex stream of information into cortical regions assumed to be
subsystems of a complex network (Cohen 2016). So, dynamically reconfigures of emotional
perception relies not only on independent processing of stimuli in particular
cortices (segregation) but also on global cooperation between these regions
(integration). Several studies show that functional brain capacity can be
measured by both segregation and integration measures such that the higher
segregation is mostly linked to simple motor tasks, while the higher
integration is mostly linked to cognitive loads (Fransson 2018; Fong 2019; Finc 2020). However, it remains a great challenge to
understand how the brain’s emotional configurations are supported by
segregation and integration measures. In experimental cognitive neuroscience,
several studies showed that the ratio between excitatory and inhibitory neural
activities remains balanced at pyramidal neurons over both time and space
(Haider 2006; Okun 2008). The more recent study reveals that
emotional brain functions have been sustained through functional connectivity
supported by a balanced ratio of excitation to inhibition originated from
neurotransmitter release in response to audio-visual and affective stimuli (Kılıç 2022). In the current study, we employed six
network measures to quantify both segregation and integration in response to
music clips over shorter time lengths.
Many neurotransmitters serve neural communications
between cortical nerve cells in order to regulate mood and emotional states.
Among them, dopamine, serotonin, endorphins, and oxytocin mediate well and
happiness. Besides, low levels of norepinephrine, serotonin, and dopamine are
associated with negative mood and unpleasant states. Apart from arousal-valance
scores of contrasting emotional states (Fear vs Anger, Happiness vs Sadness),
characteristic variations in post-synaptic potentials triggered by particular
neurotransmitters might be more distinct in between Happiness and Sadness in
comparison to Fear and Anger. Thus, this neurobiological facts might be main
factor leading to the high accuracy of 96% in classifying Happiness and Sadness
in Gamma sub-band, despite the fact that the unpleasant feelings, Fear and
Anger can be distinguished with the less CA.
According to computational and behavioral
neuroscience, affective perception is usually accompanied by changes in
high-frequency EEG series, i.e. gamma sub-bands (Boucher 2014; Aydın 2018; Yang 2020). Our results also proved that gamma-band
specific brain network measures were closely relevant to discrete emotions.
Emotion can be considered as a high-level cognitive function that requires the
re-configuration of multiple brain regions in response to external stimuli. So,
the relationships and information interactions among the cortices have been
detected in high frequency components of EEG series mediated by affective music
clips in the present study. Further, anger is found to be characterized by high
transitivity and low modularity in Gamma-band activities.
The stable neural patterns have shown across the
individuals in between negative and positive emotions induced by different
video clips of 1 min in references Zheng and Zhu (2017); Li and Liu (2019). However, Fear and Anger can not be
considered as distinct emotional states due to particular emotional labeling
principle of four quadrants assigned with low arousal/low valence, high
arousal/low valence, low arousal/high valence, and high arousal/high valence
(Zheng and Zhu 2017; Li and Liu 2019). We have used discrete emotional model to
investigate the neural dynamics underlying both Fear and Anger triggered by
musical sounds of 12 sec. Besides, the recent studies have also shown the
usefulness of Granger causality included by connectivity estimations to
classify video clips into pleasant, neutral and unpleasant quadrants of
arousal/valance dimensions (Li and Zheng 2018; Chen and Miao 2021). In more details, EEG based brain
connectivity analysis has also been successfully examined to quantify the
interactions between limbic system and motor cortex during emotional
expressions induced by video clips (Li and Li 2020). In conclusion, music clips can exactly
induce basic and discrete emotional states in very short time period such as 2
sec depending both rhythmicity and tonality of excerpt in comparison to
presentation of longer duration video clips mapped on arousal/valance
quadrants. Functional connectivity estimations can provide to insight the
hierarchial neurodynamics at both modular and system levels into EEG frequency
sub-bands.
Regarding GT based global connectivity estimations,
the important parameter is threshold that influence the resulting measures. In
analysis of fMRI data, this issue has been found to be the leading factor for
investigation of brain networks where the optimum threshold is determined
empirically between 0.2 and 0.3 (Bordier et al. 2017). Several binarization methods have been
used in combination with phase domain synchronization approaches for detection
of disorders encoded by clinical resting-state EEG recordings (Sun and Li 2019b; Tsai and Wang 2022), while two popular thresholds of the mean
and max value of individual connectivity matrix for recognition of
discrete emotional states (Kılıç 2022) and cognitive emotion regulation
strategies (Aydın 2022). The weak, noisy, and insignificant
edges/connections across the cortex are eliminated by setting a threshold in
obtaining a binary version of connectivity matrix, while the most important
connections remain. Thus, network connectivity measures are estimated from
binary adjacency matrix. Since EEG recordings are nonlinear, random, and
probabilistic time series, use of adaptive thresholds as 60% of max value and
the mean value in individual connectivity matrix computed for each short
segment for classification of discrete emotional states induced by musical
sounds in the present study.