Detrending Moving Average, Power Spectral Density, and Coherence: Three EEG-Based Methods to Assess Emotion Irradiation during Facial Perception

Round 1

Reviewer 1 Report

The presented paper has interesting motivation to investigate neural correlates during to assess emotion during face perception. Although the data set and the experiment seem robust for the goal, the performed analysis could be improved.

Here are some overall questions:

* Why divide the brain rhythms into so many sub-divisions? There is some literature that could support this division, but in dealing with EEG, the information can be tricky and is not the best approach to go with.

 

* There is a lack o clarity on the decisions made over the statistical procedures. For all the multielectrode analyses, seeking general patterns, I suggest using permutation analysis.

 

* Indeed functional connectivity analysis is based on the study of neural dynamics through some metrics that evaluate the correlation of activity in different brain regions/scales. The problem is that in your paper is a mere correlation in the coherence between different electrodes. There is no robust network analysis, so building a discussion over the network effects is not the right direction given your results. Even more, given connectivity studies with EEG fell short due to a lack of spatial resolution, which in a lot of cases is solved by using source localization (when is possible to apply).

* The figures could use a better organization that helps to understand which are the main results. 

 

* There are some missing references in the introduction.

 

* PSD does not highlight synchronization/desynchronization effects. 

 

* In line 303 there is a missing reference.

 

* Consider rewriting some of the points on the discussion considering that the EEG does not have the spatial resolution to address comments on specific structures as fusiform gyrus, for example.

 

 

 

Author Response

* Why divide the brain rhythms into so many sub-divisions? There is some literature that could support this division, but in dealing with EEG, the information can be tricky and is not the best approach to go with.

Thank you for your remark, however, our approach, in this case, was based on the traditional view of the heterogeneous nature of different subbands of the EEG activity, both in terms of their origin and functional role. These are some of the papers which describe the major reasoning underlying such a concept:

Petsche H., Kaplan S., von Stein A., Filz O. The possible meaning of the upper and lower alpha frequency ranges for cognitive and creative tasks // Int. J. Psychophysiol.   1997. V. 26.N1-3. P.77.

K. Engel and P. Fries, “Beta-band oscillations-signalling the status quo?” Current Opin. Neurobiol. 2010, 20, No. 2, 156-165 . DOI: 10.1016/j.conb.2010.02.015

Kukleta, M. Brázdil, R. Roman, et al., “Cognitive network  interactions and beta-2 coherence in processing non-target stimuli in visual oddball task,” Physiol. Res. 2009, 58, 139-148 . DOI: 10.33549/physiolres.931404

* There is a lack o clarity on the decisions made over the statistical procedures. For all the multielectrode analyses, seeking general patterns, I suggest using permutation analysis.

Thank you for your remark. For this particular study, we followed conventional statistical procedures, that are used in the field of neuroscience - comparing the distributions between two pairs of electrodes in different experimental conditions. For now, it is not clear to us how we could use the permutation analysis is this experiment. However, we would definitely consider using it, if we have a suitable experiment condition.

Indeed functional connectivity analysis is based on the study of neural dynamics through some metrics that evaluate the correlation of activity in different brain regions/scales. The problem is that in your paper is a mere correlation in the coherence between different electrodes. There is no robust network analysis, so building a discussion over the network effects is not the right direction given your results. Even more, given connectivity studies with EEG fell short due to a lack of spatial resolution, which in a lot of cases is solved by using source localization (when is possible to apply).

Thank you for putting emphasis on this issue, and we have to partially agree with your remark. Bearing this in mind, we have made changes to the discussion section, removing the points that mention network analysis in favor of correlation description.  The spatial resolution might seem relatively low, compared to the modern EEG-headsets. However, we believe that this density of EEG array is rather sufficient to make conclusions about general neurophysiological patterns, as it includes crucial cortical regions, which can be seen as basic strong points of neural ensembles.

The figures could use a better organization that helps to understand which are the main results. 

Thank you for pointing this out. We wanted to highlight all changes on one page according to each section in the most logical way. Figures were tried to make clear with high resolution to show all the information correctly.

There are some missing references in the introduction.

 We really appreciate that you’ve pointed this out. References were added to the the introduction part of the paper.

PSD does not highlight synchronization/desynchronization effects. 

Thank you for your note, which we agree with. The expression is a bit outdated and not strict. In order to make it more precise, we’ve replaced it with “ {an increase or decrease in the power of the particular oscillation range” (lines 35,36).

In line 303 there is a missing reference.

Thank you for pointing this out. We’ve added [Pomper U. ,  Ansorge U.  Theta-Rhythmic Oscillation of Working Memory Performance. Psychol Sci. 2021;32(11):1801-1810 . doi: 0.1177/09567976211013045] instead of the missing reference. 

Consider rewriting some of the points on the discussion considering that the EEG does not have the spatial resolution to address comments on specific structures as fusiform gyrus, for example.

We extremely appreciate your suggestion, and we absolutely agree with it. Therefore, changes have been made to some parts of the discussion part so that direct references to the brain structures were omitted.

Reviewer 2 Report

Paper presents basic analyses of the EEG differences in the response to presented neutral and emotional face stimuli. Face perception and emotion analyses with EEG are quite difficult and usually require more sophisticated methods. I have several comments to this work:
General comments:
1. cited literature is quite old (on average over 10 years)
2. the computational methods are very basic
3. I did not find direct comparison of the applied methods as suggested in the introduction.
In consequence obtained results are at best confirmatory.
I have also some methodological concerns:
1. density of EEG array is low
2. spectral coherence used for connectivity assessment is a measure which is subject to the error associated with tissue conductivity, therefore more reliable ways to estimate connectivity would be those using phase differences or phase correlations (there are quite a few available).
3. the coherence comparisons between two conditions are not based on their direct differences  which makes it difficult.
 I would suggest to apply more sophisticated analytical methods such as basic machine learning tools such as SVM, random forest which could possibly show some differences overlooked using basic methods

Author Response

1. cited literature is quite old (on average over 10 years)
   We are grateful for your remark regarding the list of the cited literature. We’ve revised it and added more recently published works.
   
2. the computational methods are very basic.
   

   Thank you for noting that in this work we used basic methods, with which we partially agree. For sure, both spectral analysis and coherence analysis of EEG are classical techniques that have proven their effectiveness for many years and the majority of tasks. Our motivation to use them in our task is twofold. First, we would like to apply PSD and coherence analysis to the considered task, since the interpretation of the results is clear and will be accepted by the EEG community, while to our knowledge this is the first usage of these methods for the experimental setting used in our work (comparison of groups, sequences). We believe that PSD and coherence in such experiments gave useful insights into brain activity. Second, our aim is to relate the PSD and coherence information with the characteristics provided by the detrended moving average analysis of scaling properties of EEG, which is comparatively new in the EEG analysis domain. We used these three methods together to analyze three different features of EEG: strength of oscillations (PSD), the connection between different regions (coherence), and scaling characteristics (DMA).

   
   

3. I did not find direct comparison of the applied methods as suggested in the introduction. In consequence obtained results are at best confirmatory.
   Thank you for your comment. We did not set the specific goal of a physical comparison of the applied methods (PSD, Coh, DMA). Each of them has its own advantages and disadvantages. Therefore, the introduction focuses mainly on describing the signal parameters that underlie each method. Our goal was to find a set of procedures that, when combined, can provide a picture of the neuromechanisms of the brain under certain conditions (for example, as in neuroimaging, the combined use of CT, PET, and MRI provides the clearest diagnostic results). We have also slightly modified the introduction, so that emphasis was put primarily on a combination of methods.

4. density of EEG array is low
   
   Thank you for your remark, which we can partially agree with. From one point of view, the number of channels might seem relatively low, compared to the modern EEG-headsets. However, we believe that this density of EEG array is rather sufficient to make conclusions about general neurophysiological patterns, as it includes crucial cortical regions, which can be seen as basic strong points of neural ensembles.
   
5. spectral coherence used for connectivity assessment is a measure which is subject to the error associated with tissue conductivity, therefore more reliable ways to estimate connectivity would be those using phase differences or phase correlations (there are quite a few available)
   
   

   We are really grateful for your suggestion and must admit that the coherence measurement method has been a subject of discussion for some time. For the purposes of this paper, we've decided to use coherence as a basic method of connectivity assessment, aiming at further development of this research with a wider range of methodological approaches, including such methods as phase transfer entropy measurement and dynamic causal modeling.

6. the coherence comparisons between two conditions are not based on their direct differences  which makes it difficult.  I would suggest to apply more sophisticated analytical methods such as basic machine learning tools such as SVM, random forest which could possibly show some differences overlooked using basic methods.
   
   

   Thank you for your comment. Indeed it was possible to overlook some differences between conditions if you consider the classification task or the problem of quantitative differentiation between mental states. However, in this paper, we would like to point out the differences between states using the conventional techniques that are used in the field of neuroscience, such as PSD and coherence, together with the DMA in order to qualitatively distinguish the difference between mental states rather from the neuroscientific point of view. Of course, your suggestion would be a great continuation of this paper. 

    

Reviewer 3 Report

It is a well-written manuscript that studied the power spectral density, detrended moving average, and coherence analysis of the EEG signals. The authors reported that the theta and beta networks are actively different with gender.

The following comments should be considered in the revised manuscript:

 

1) The subject's informed consent and information about the experimental protocol reviewed by ethics committee are also need to provided in Section 2.1.

2) The detailed frequency range of the bandpass filter should be added in Section 2.4, or providing the frequency-magnitude response plot instead.

3) Figure 5 can be revised to extend the display area, for example, the subfigure A may occur an entire line, then subfigure B is shown below, and so forth.

4) Most of the references are out of date. Please provide more literature published in the recent 5 years.

Author Response

1) The subject's informed consent and information about the experimental protocol reviewed by ethics committee are also need to provided in Section 2.1.

Thank you for pointing this out. We’ve added some additional information regarding this issue in lines 54-61.

2) The detailed frequency range of the bandpass filter should be added in Section 2.4, or providing the frequency-magnitude response plot instead.

We have added the information with specification of cutoff frequencies for Bandpass filter. In our research we used only classical ways of creating filter according to the basic description of Python libraries (e.g. https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.butter.html, https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.lfilter.html).

3) Figure 5 can be revised to extend the display area, for example, the subfigure A may occur an entire line, then subfigure B is shown below, and so forth.

Thank you for pointing this out. For clear visualization of coherence statistical tests’ results, we decided to use the 2x2 format for presenting the figures. We wanted to highlight all changes on one page, so our variant is the most comfortable for observing results fully according to the template of the journal. We tried to make the figures clear with high resolution to show all information correctly.

4) Most of the references are out of date. Please provide more literature published in the recent 5 years.

We are very grateful for your consideration. Our list of cited literature has been revised and updated.

Round 2

Reviewer 1 Report

* The first question made previously remains the same. There is literature about the sub-divisions of the brain rhythms, but some of them used in this paper you didn't cite the source for it. So it seems a bit too explorative without a focused goal (vide alpha 3, theta 1/2). There is an absence of a clear reason for why divide these bands, particularly on theta, since later on you only discuss theta rhythm in general. The papers that you send me don't support all the divisions that you have chosen.

 

* In dealing with multielectrode analysis, by doing just a pairwise comparison some false positives can be found due to multicomparison of electrodes. Given that the EEG does not have the best spatial resolution a cluster permutation analysis can be useful to also check the effect considering the null distribution of other electrodes.

 

Author Response

1. We are really grateful for your comment. However, as we pointed out in our previous replies, in our work, we were guided by the concepts of the physiological and functional heterogeneity of the brain's rhythmic activity, which were developed previously. In the context of these ideas, all classical EEG bands can be divided into subranges, which makes it possible to study the mechanisms of cognitive activity of the brain more subtly. 

For instance, researchers consider the theta-1 subband (4-5 Hz) to be a correlate of emotional activity, and theta-2 is found to be the correlate of the memory-related processes.

Klimesch W., Doppelmayr M., Russegger H., Pachinger T., Schwaiger J. Induced alpha band power in the human EEG and attention // Neyrosc. Letters. 1998. V.244. P.73-76.

Rohm D., Klimesch W., Haider H., Doppelmayr M. The role of theta and alpha oscillations for language comprehension in the human electroencephalogram // Neurosci. Lett. 2001.Vol.310,No The theta-17-140.

Theta-1 range is found to be sensitive to arithmetic problems solving, and beta-2 is demonstrative in case of heuristic task execution.

Razoumnikova O.M. Functional organization of different brain areas during convergent and divergent thinking: An EEG investigation // Cogn. Brain Res. 2000. №10. P. 11–18.

Razoumnikova O.M. Gender differences in hemispheric organization during divergent thinking: An EEG investigation in human subjects // Neurosci. Lett. 2004. V. 362. №3. к. 193–195.

I V Tarasova , N V Volf, O M Razumnikova Changes in the coherence of cortical biopotentials during performance of a verbal creative task in men and women

Teplan M., Krakovská A., Štolc S.  EEG responses to long-term audio–visual stimulation Int. J.  Psychophysiology. 2006. Vol.59, Issue 2. P.81-90 DOI: 10.1016/j.ijpsycho.2005.02.005

apart from that, theta-2 is seen as a subband of the rhythmic brain activity, which reflects the actualization of the cognitive component of emotional response.

An increase in theta1 power - in the temporal, parietal, and occipital regions is found to reflect the functioning of motivational attention mechanisms, which is accompanied by the activation of the information accessing and retrieving mechanisms related to previous emotional experiences.

L I Aftanas 1, A A Varlamov, N V Reva, S V Pavlov Disruption of early event-related theta synchronization of human EEG in alexithymics viewing affective pictures // Neurosci Lett . 2003;340(1):57-60. doi: 10.1016/s0304-3940(03)00070-3.

As for the alpha EEG-subband, according to the literary data, alpha-1 (8-9 Hz) has an inherent role as a correlate of memory processes; alpha2- (10-11 Hz) – as the marker of attention-related processes; alpha3- (12-13 Hz) is associated with the processes of inner speech, logical thinking.

Vogt F., Klimesch W., Doppelmayr M. High-frequency components in the alpha band and memory performance // J. Clin. Neurophysiol. 1998. Vol.15,No2. P.167-172.

Feshchenko V.A., Reinsel R.A., Veselis R.A. Multiplicity of the alpha fhythm in normal humans // J.Clin. Neurophysiol. 2001. Vol.18,No4. P.331-344.

Sepideh Sadaghiani, Rene' Scheeringa, Katia Lehongre, Benjamin Morillon, Anne-Lise Giraud, and Andreas Kleinschmidt   Intrinsic Connectivity Networks, Alpha Oscillations, and Tonic Alertness: A Simultaneous Electroencephalography/ Functional Magnetic Resonance Imaging Study //  J. Neurosci.,  2010 004-10.2010             DOI:10.1523/JNEUROSCI.1004-10.2010  

Aftanas L.I., Golocheikine S.A. Human anterior and frontal midline theta and lower alpha reflect     emotionally positive state and internalized attention:     high-resolution EEG investigation of meditation // Neurosci Lett. 2001. Vol.310,No1. P.57-60.

Synchronization in alpha-3 subband is thought to reflect the activation of episodic short-term memory, which reflects the mechanism of the long-term semantic memory inhibition - thereby being a kind of filter to enhance the signal-to-noise ratio for the subsequent selective activation of semantic processes. A decrease in the alpha-3 activity is associated with processes of semantic long-term memory.

Klimesch W., Doppelmayr M., Russegger H., Pachinger T., Schwaiger J. Induced alpha band power in the human EEG and attention // Neyrosc. Letters. 1998. V.244. P.73-76.

Klimesch W.,  Doppelmayr M., Schwaiger J., Auinger P., Winkler Th. `Paradoxical` alpha synchronization in a memory task // Cognitive Brain Research 1999. V.7. P.493-501.

Desynchronization of low-frequency alpha is partly related to attention-connected mechanisms. 

Klimesch W.,  Doppelmayr M., Rohm D., Pollhuber D., Stadler W. Simultaneous desynchronization and  synchronization of different alpha responses in the human electroencephalograph: a neglected paradox? // Neuroscience Letters 2000. V.284. P.97-100.

At the same time, increased alpha-3 activity in the frontal cortical areas is observed during short-term memory task performance, which is known to activate the frontal regions of the neocortex.

Goldman-Rakic P.S. Regional and cellular fractionation of working memory Proc. Natl. Acad. Sci. 1996. V.93. P.13473-13480

During the improvement of the mood, a decrease in COG 9.5-12 Hz and an increase in amplitude 7.5-9.0 were observed

Petsche H. Kaplan S., von Stein A., Filz O. The possible meaning of the upper and lower alpha frequency ranges     for cognitive and creative tasks // Int. J. Psychophysiol. 1997.Vol.26, No1-3. P.77-97

Lastly, we would like to illustrate our subband selection with a study, which showed that repetitive training with audio–visual stimulation induces changes in the brain's electrocortical activity. Long-term AVS exposure significantly increases power in theta-1, theta-2, and alpha-1 bands in the frontal and central cortex locations, and total power increases in the right central region. Interhemispheric coherence in the alpha-1 band displayed a significant increase between frontal parts.

Teplan M., Krakovská A., Štolc S.  EEG responses to long-term audio–visual stimulation Int. J.  Psychophysiology. 2006. Vol.59, Issue 2. P.81-90

At the same time, the upper alpha band (10 –12 Hz) oscillations is known as the most consistent electrical correlate of tonic alertness.

Sepideh Sadaghiani, Rene' Scheeringa, Katia Lehongre, Benjamin Morillon, Anne-Lise Giraud, and Andreas Kleinschmidt  Intrinsic Connectivity Networks, Alpha Oscillations, and Tonic Alertness: A Simultaneous Electroencephalography/ Functional Magnetic Resonance Imaging Study // The Journal of Neuroscience, 2010 • 30(30):10243–10250  DOI:10.1523/JNEUROSCI.1004-10.2010

In our opinion, the above list of essentially classical works proves the need for such an approach to EEG data analysis. In our work, we discussed only those results supported by the statistical analysis data. The work is essentially a search; it was important for us to describe the phenomenology of changes in EEG parameters brought up by different processing methods.

2. In our opinion, this remark is more accurate concerning the case of a 10-10% electrode application system or high-density EEG array.

Reviewer 2 Report

Paper is neither methodological nor physiological. It compares well known methods using well known paradigm. Authors show a lot of results without selecting those of highest importance for them.

Author Response

Thank you for your comment. Our goal was to show the nuances of the objective interpretation of the detrended moving algorithm application compared to more widely known and used methods, providing a general picture. As in the case of neuroimaging, a complete picture is obtained with the combination of CT, MRI, and PET data analysis. The purpose of the work was to conduct a comparative analysis of the methods with different mathematical apparatus for their further most accurate application in subsequent research using other stimulus models.