Neuromagnetism “On the Cheap”: Evaluating a Combined Cylindrical Shield and Partial-Coverage OPM-MEG System for Detecting Sensorimotor Responses in Humans

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The authors describe a "low-cost" OPM-MEG system that can hopefully increase the accessibility of OPM-MEG to labs with lower sources of funding or just starting out in MEG. The results presented look promising and convincing and it is encouraging to see such results are achievable even without active shielding. While the work is not particularly novel in my opinion (most of this was shown in the "early" days of OPM-MEG), it is still perhaps worthwhile to repeat the measures with newer techniques in place (such as HFC) and thus will be of some interest to other readers. 

My main concern with this paper is the use of HFC for such a small sensor count over a small area is not validated. In my recent anecdotal experience of performing phantom measurements with a 16-channel array, I found that using HFC actuallly supressed a lot of the phantom activitiy because the sensors being in such close proximity to one another all "saw" the same field. I wonder if this is potentially what is happening with the lower SNR than other reported studies? I think it would be worth seeing a comparison of the data both in sensor and source space before and after HFC. While the sensor space activity may look noisier, source localisation often does just as good a job at cleaning the data at source level. The two OPM references (Boto and Borna) that the SNR is compared against both did not use HFC so the comparison is potentially flawed. 

Additionally, there is no actual cost comparison. It would be good to know at least to an order of magnitude what the cost of this system is compared to the full systems. Any reader interested in this work is likely looking into how they can get started and cost information is important. 

Minor points:

1) Line 381 - Typo: "not eliminated"
2) Is the use of a BEM overkill when you are already using a template head? Could that have also have negative effects on SNR if the underlying source is already somewhat incorrect?

 

Author Response

Dear Editor,

Thank you for giving us the opportunity to revise and resubmit our manuscript. We have updated the manuscript in response to reviewer feedback; below we have provided direct responses to each reviewer comment. Reviewer comments are numbered and printed in black ink; our responses to each are printed in blue ink. All references to in-text changes are indicated by line number in the updated manuscript, and are printed in blue and italicized below. All changes within the manuscript itself are highlighted for easy identification.  

 

 

 

Responses to Reviewer 1

• The authors describe a "low-cost" OPM-MEG system that can hopefully increase the accessibility of OPM-MEG to labs with lower sources of funding or just starting out in MEG. The results presented look promising and convincing and it is encouraging to see such results are achievable even without active shielding. While the work is not particularly novel in my opinion (most of this was shown in the "early" days of OPM-MEG), it is still perhaps worthwhile to repeat the measures with newer techniques in place (such as HFC) and thus will be of some interest to other readers. 

 

We thank the reviewer for their kind evaluation.

 

 

• My main concern with this paper is the use of HFC for such a small sensor count over a small area is not validated. In my recent anecdotal experience of performing phantom measurements with a 16-channel array, I found that using HFC actuallly supressed a lot of the phantom activitiy because the sensors being in such close proximity to one another all "saw" the same field. I wonder if this is potentially what is happening with the lower SNR than other reported studies? I think it would be worth seeing a comparison of the data both in sensor and source space before and after HFC. While the sensor space activity may look noisier, source localisation often does just as good a job at cleaning the data at source level. The two OPM references (Boto and Borna) that the SNR is compared against both did not use HFC so the comparison is potentially flawed. 

We thank the reviewer for this valuable suggestion. We replicated our analyses without HFC and found that HFC removes some signal magnitude from the evoked field, but still improves our SNR. The improvement in SNR is apparent at sensor and source levels. Finally, the changes in mu and beta cannot be resolved at the sensor level without HFC (i.e., HFC is required to see sensor level changes in mu/beta). We have added the following content to the manuscript:

[Lines 485-504] "HFC is typically applied to arrays with high channel counts and whole-head coverage; to our knowledge HFC has not been validated in partial-coverage arrays with low sensor counts. We therefore conducted a supplementary analysis to assess whether this step was appropriate for our data. We repeated our original analyses but this time omitted HFC during preprocessing. To maintain consistency between the two sets of analyses (with and without HFC), we removed the same manually identified bad channels and epochs during preprocessing. We compared results obtained with versus without HFC in terms of response magnitudes and SNR; quantitative outcomes of this comparison are briefly described here, with detailed results in Supplementary Materials. At the sensor level, applying HFC dramatically reduced the magnitude of SEF responses and movement-related mu/beta modulation. These effects were less pronounced at the source level: we saw a slight reduction in SEF magnitude across all ROIs, while oscillatory (mu/beta) responses appeared unaffected. Despite the apparent attenuation of neural responses induced by HFC, we found that performing analyses without HFC generally reduced SNR at both the sensor and source levels. For example, skipping HFC reduced the maximum SNR across ROIs by ~65% (median across subjects), and also shifted the location (channel/ROI) at which SNR was maximal towards posterior areas - away from central areas where the SEF is typically observed. These results suggest that HFC was beneficial and removed relatively more noise from the recordings than true signal."

[Lines 552-555] "We note that performing HFC (not applied in previous work with comparable partial-coverage systems [4,8]) did improve SNR overall, despite our sensors being arranged over a relatively small area, indicating that this preprocessing step can provide at least marginal gains when working with partial-coverage arrays."

Lines [602-604] "Moreover, modern preprocessing methods such as HFC, previously applied to data from systems with whole-head coverage, can improve SNR in partial-coverage systems such as this one."

We have also provided Supplementary Materials detailing results obtained without HFC. Specifically, we provide a table comparing our SNR metrics between HFC ON and HFC OFF, and we provide a figure of main results with HFC OFF (for comparison to figures in the main body).

• Additionally, there is no actual cost comparison. It would be good to know at least to an order of magnitude what the cost of this system is compared to the full systems. Any reader interested in this work is likely looking into how they can get started and cost information is important. 

We thank the reviewer for this suggestion. While we see the appeal of a direct cost comparison, we are concerned that any explicit $ values we provide will quickly become outdated, either due to inflation, industry changes in production efficiency, tariffs, etc. Therefore, we have instead provided a comparison of mu-metal square-footage demands, which are a major factor contributing the price tag of any system:

[Lines 615-623] "A major benefit of the system described here is that it requires considerably less mu-metal than one housed in an MSR. Mu-metal is both financially and environmentally costly, making up a large percentage of the purchase price of MEG systems. The cylindrical shield requires a surface area of around 12-13m² of mu-metal (2 walls, ~0.9m diameter and ~2m length, plus the end caps and a reducer). By contrast, a state-of-the-art "lightweight" MSR [46] requires around 84m² of shielding³. Therefore, a compact system such as ours offers a roughly six-fold reduction in mu-metal materials".

Footnote: ³ Mu-metal square footage was estimated based on inner and outer volumes: 2.4 x 2.4 x 2.4m and 2.8 x 3 x 2.8m, respectively, reported in [46]"

We have also

 

• Line 381 - Typo: "not eliminated"

Thank you. We have corrected this.

1.5.        Is the use of a BEM overkill when you are already using a template head? Could that have also have negative effects on SNR if the underlying source is already somewhat incorrect?

We used BEM because (in our experience) it is standard for co-registration prior to source localization with minimum norm estimation. It is possible that the scaling of the fsaverage BEM surfaces (that is, to each participant's head) induced some error; however, some degree of coregistration error is to be expected in any pipeline, even when using participant-specific MRIs. Moreover, BEMs are commonplace for source localization, and we feel it is worthwhile presenting results obtained from a typical / generalizable pipeline.  Note that we are not using the source estimation results to suggest source location at a specific vertex (where the reviewer’s concern would have a larger impact. Instead, we are reporting mean activity within each ROI using the “PCA mean” of all vertices within this ROI, where we expect that any negative effect of BEM inaccuracy is averaged out and less relevant.

 

 

Responses to Reviewer 2

2.1.        This is an interesting paper that examines measuring the magnetoencephalogram       using optical magnetometers and a cylindrical shield. This is important for reducing              the cost and complexity of using MEG in many clinical settings.

We thank the reviewer for this kind evaluation.

 

2.2.        The authors describe the technical details of their methods quite well. But they              don’t tell us about the experience from the point of view of the patient. The          experiments are performed in a narrow cylinder with the sensors in a fixed helmet,      which is a very unnatural environment. I would expect that the patients would be             uncomfortable inside the coffin-like cylindrical shield. Could any of the differences               between the author’s results and other results reported in the literature be        explained by the patient being under stress?

We think it extremely unlikely that our participants experienced levels of stress beyond the typical discomforts of any neuroimaging study. As detailed in our responses below, we took great care to ensure the comfort of our participants, and none reported any distress or discomfort, even when asked directly about their experience inside the scanner during debriefing. Importantly, we highlight that the internal radius of the shield (43 cm) is wider than the bore of a typical MRI scanner, where fMRI experiments are regularly performed. 

 

2.3.        Were the patients given time to get used to the shielded room before taking data?

Yes. We have updated the text to emphasize this: [Lines 174-176] "Each participant completed informed consent and was given a brief tour of the OPM system before beginning the experiment, and was encouraged to ask any questions or express any concerns about the experiment before entering the scanner."

 

2.4.        How far were the patients actually pushed into the cylinder? For instance, were they pushed in so far that their feet were inside, or just pushed in up to their shoulders                 like in Fig. 1? ... Was there any lighting inside the cylinder? What was the              temperature in the cylinder? Is the air circulation inside the cylinder adequate to      maintain a normal oxygen and carbon dioxide concentration? ... The article                   mentioned friction-fitted end caps. Does this mean that the patient was completely enclosed so they had no exposure to the outside at all?

We have updated the text with the following: [Lines 136-144] "The cylinder is mounted on a wooden cradle, and participants are supported (lying supine) on a retractable wooden bed, on a non-metal rail system which allows the participant to be inserted head-first until the helmet is approximately 1.65m into the cylinder. The shield end at the participant’s feet always remains open (i.e., there is an end cap at the participant’s head end only). Once inserted into the shield, the participant’s head is situated about 60cm from the end cap. A small bore in the end-cap (enabling external projection of visual stimuli; not used here) allows air flow through the cylinder (no heating or cooling is required). Ambient lighting from the surrounding room allows participants to see inside the cylinder. "

 

2.5.        Did the rigid helmet fit each patient equally well? Did it press or squeeze the head        for any patients?

We have added the following text: [Lines 181-184] "Layers of foam padding were used to cushion the back of the participant’s head against the helmet and, when necessary, additional padding was inserted to ensure that the participant was as comfortable as possible and to reduce head movement inside the helmet."

 

2.6.        Was it quiet in the cylinder, or were there any sources of noise (I don’t know if these    optical sensors are silent)?

We have added the following: [Lines 161-163] "The sensors are silent and operate at a maximum temperature of ~30°C, meaning they can be placed safely and comfortably against the head."

 

2.7.        How long were patients typically inside the cylinder?

We have added the following: [Lines 191-192] "Each participant was inside the scanner for approximately 30 minutes".

 

2.8.        Did any complain of headaches, anxiety, or claustrophobia afterwards?

No. We have added the following text: [120-123] "During recruitment and informed consent, potential participants were advised that individuals with claustrophobia might experience anxiety inside the shield, and that they should consider this before agreeing to participate." And [195-196] "After exiting the scanner, each participant was asked about their experience of the experiment. None reported excessive discomfort."

 

2.9.        Did any patient refuse to complete the study because they were not comfortable          inside the cylinder? ... Could the researchers communicate with the patient during             the experiment?

No participants withdrew from the study. We have updated the text with the following: [Lines 192-194] "Between tasks, a member of the research team communicated verbally with the participant to inform them of the upcoming task, and to ensure that they were still comfortable inside the scanner."

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This is an interesting paper that examines measuring the magnetoencephalogram using optical magnetometers and a cylindrical shield. This is important for reducing the cost and complexity of using MEG in many clinical settings.

The authors describe the technical details of their methods quite well. But they don’t tell us about the experience from the point of view of the patient. The experiments are performed in a narrow cylinder with the sensors in a fixed helmet, which is a very unnatural environment. I would expect that the patients would be uncomfortable inside the coffin-like cylindrical shield. Could any of the differences between the author’s results and other results reported in the literature be explained by the patient being under stress? Were the patients given time to get used to the shielded room before taking data? How far were the patients actually pushed into the cylinder? For instance, were they pushed in so far that their feet were inside, or just pushed in up to their shoulders like in Fig. 1? Did the rigid helmet fit each patient equally well? Did it press or squeeze the head for any patients? Was there any lighting inside the cylinder? What was the temperature in the cylinder? Is the air circulation inside the cylinder adequate to maintain a normal oxygen and carbon dioxide concentration? Was it quiet in the cylinder, or were there any sources of noise (I don’t know if these optical sensors are silent)? How long were patients typically inside the cylinder? Did any complain of headaches, anxiety, or claustrophobia afterwards? Did any patient refuse to complete the study because they were not comfortable inside the cylinder? The article mentioned friction-fitted end caps. Does this mean that the patient was completely enclosed so they had no exposure to the outside at all? Could the researchers communicate with the patient during the experiment?

Author Response

Dear Editor,

Thank you for giving us the opportunity to revise and resubmit our manuscript. We have updated the manuscript in response to reviewer feedback; below we have provided direct responses to each reviewer comment. Reviewer comments are numbered and printed in black ink; our responses to each are printed in blue ink. All references to in-text changes are indicated by line number in the updated manuscript, and are printed in blue and italicized below. All changes within the manuscript itself are highlighted for easy identification.  

 

 

 

Responses to Reviewer 1

• The authors describe a "low-cost" OPM-MEG system that can hopefully increase the accessibility of OPM-MEG to labs with lower sources of funding or just starting out in MEG. The results presented look promising and convincing and it is encouraging to see such results are achievable even without active shielding. While the work is not particularly novel in my opinion (most of this was shown in the "early" days of OPM-MEG), it is still perhaps worthwhile to repeat the measures with newer techniques in place (such as HFC) and thus will be of some interest to other readers. 

 

We thank the reviewer for their kind evaluation.

 

 

• My main concern with this paper is the use of HFC for such a small sensor count over a small area is not validated. In my recent anecdotal experience of performing phantom measurements with a 16-channel array, I found that using HFC actuallly supressed a lot of the phantom activitiy because the sensors being in such close proximity to one another all "saw" the same field. I wonder if this is potentially what is happening with the lower SNR than other reported studies? I think it would be worth seeing a comparison of the data both in sensor and source space before and after HFC. While the sensor space activity may look noisier, source localisation often does just as good a job at cleaning the data at source level. The two OPM references (Boto and Borna) that the SNR is compared against both did not use HFC so the comparison is potentially flawed. 

We thank the reviewer for this valuable suggestion. We replicated our analyses without HFC and found that HFC removes some signal magnitude from the evoked field, but still improves our SNR. The improvement in SNR is apparent at sensor and source levels. Finally, the changes in mu and beta cannot be resolved at the sensor level without HFC (i.e., HFC is required to see sensor level changes in mu/beta). We have added the following content to the manuscript:

[Lines 485-504] "HFC is typically applied to arrays with high channel counts and whole-head coverage; to our knowledge HFC has not been validated in partial-coverage arrays with low sensor counts. We therefore conducted a supplementary analysis to assess whether this step was appropriate for our data. We repeated our original analyses but this time omitted HFC during preprocessing. To maintain consistency between the two sets of analyses (with and without HFC), we removed the same manually identified bad channels and epochs during preprocessing. We compared results obtained with versus without HFC in terms of response magnitudes and SNR; quantitative outcomes of this comparison are briefly described here, with detailed results in Supplementary Materials. At the sensor level, applying HFC dramatically reduced the magnitude of SEF responses and movement-related mu/beta modulation. These effects were less pronounced at the source level: we saw a slight reduction in SEF magnitude across all ROIs, while oscillatory (mu/beta) responses appeared unaffected. Despite the apparent attenuation of neural responses induced by HFC, we found that performing analyses without HFC generally reduced SNR at both the sensor and source levels. For example, skipping HFC reduced the maximum SNR across ROIs by ~65% (median across subjects), and also shifted the location (channel/ROI) at which SNR was maximal towards posterior areas - away from central areas where the SEF is typically observed. These results suggest that HFC was beneficial and removed relatively more noise from the recordings than true signal."

[Lines 552-555] "We note that performing HFC (not applied in previous work with comparable partial-coverage systems [4,8]) did improve SNR overall, despite our sensors being arranged over a relatively small area, indicating that this preprocessing step can provide at least marginal gains when working with partial-coverage arrays."

Lines [602-604] "Moreover, modern preprocessing methods such as HFC, previously applied to data from systems with whole-head coverage, can improve SNR in partial-coverage systems such as this one."

We have also provided Supplementary Materials detailing results obtained without HFC. Specifically, we provide a table comparing our SNR metrics between HFC ON and HFC OFF, and we provide a figure of main results with HFC OFF (for comparison to figures in the main body).

• Additionally, there is no actual cost comparison. It would be good to know at least to an order of magnitude what the cost of this system is compared to the full systems. Any reader interested in this work is likely looking into how they can get started and cost information is important. 

We thank the reviewer for this suggestion. While we see the appeal of a direct cost comparison, we are concerned that any explicit $ values we provide will quickly become outdated, either due to inflation, industry changes in production efficiency, tariffs, etc. Therefore, we have instead provided a comparison of mu-metal square-footage demands, which are a major factor contributing the price tag of any system:

[Lines 615-623] "A major benefit of the system described here is that it requires considerably less mu-metal than one housed in an MSR. Mu-metal is both financially and environmentally costly, making up a large percentage of the purchase price of MEG systems. The cylindrical shield requires a surface area of around 12-13m² of mu-metal (2 walls, ~0.9m diameter and ~2m length, plus the end caps and a reducer). By contrast, a state-of-the-art "lightweight" MSR [46] requires around 84m² of shielding³. Therefore, a compact system such as ours offers a roughly six-fold reduction in mu-metal materials".

Footnote: ³ Mu-metal square footage was estimated based on inner and outer volumes: 2.4 x 2.4 x 2.4m and 2.8 x 3 x 2.8m, respectively, reported in [46]"

We have also

 

• Line 381 - Typo: "not eliminated"

Thank you. We have corrected this.

1.5.        Is the use of a BEM overkill when you are already using a template head? Could that have also have negative effects on SNR if the underlying source is already somewhat incorrect?

We used BEM because (in our experience) it is standard for co-registration prior to source localization with minimum norm estimation. It is possible that the scaling of the fsaverage BEM surfaces (that is, to each participant's head) induced some error; however, some degree of coregistration error is to be expected in any pipeline, even when using participant-specific MRIs. Moreover, BEMs are commonplace for source localization, and we feel it is worthwhile presenting results obtained from a typical / generalizable pipeline.  Note that we are not using the source estimation results to suggest source location at a specific vertex (where the reviewer’s concern would have a larger impact. Instead, we are reporting mean activity within each ROI using the “PCA mean” of all vertices within this ROI, where we expect that any negative effect of BEM inaccuracy is averaged out and less relevant.

 

 

Responses to Reviewer 2

2.1.        This is an interesting paper that examines measuring the magnetoencephalogram       using optical magnetometers and a cylindrical shield. This is important for reducing              the cost and complexity of using MEG in many clinical settings.

We thank the reviewer for this kind evaluation.

 

2.2.        The authors describe the technical details of their methods quite well. But they              don’t tell us about the experience from the point of view of the patient. The          experiments are performed in a narrow cylinder with the sensors in a fixed helmet,      which is a very unnatural environment. I would expect that the patients would be             uncomfortable inside the coffin-like cylindrical shield. Could any of the differences               between the author’s results and other results reported in the literature be        explained by the patient being under stress?

We think it extremely unlikely that our participants experienced levels of stress beyond the typical discomforts of any neuroimaging study. As detailed in our responses below, we took great care to ensure the comfort of our participants, and none reported any distress or discomfort, even when asked directly about their experience inside the scanner during debriefing. Importantly, we highlight that the internal radius of the shield (43 cm) is wider than the bore of a typical MRI scanner, where fMRI experiments are regularly performed. 

 

2.3.        Were the patients given time to get used to the shielded room before taking data?

Yes. We have updated the text to emphasize this: [Lines 174-176] "Each participant completed informed consent and was given a brief tour of the OPM system before beginning the experiment, and was encouraged to ask any questions or express any concerns about the experiment before entering the scanner."

 

2.4.        How far were the patients actually pushed into the cylinder? For instance, were they pushed in so far that their feet were inside, or just pushed in up to their shoulders                 like in Fig. 1? ... Was there any lighting inside the cylinder? What was the              temperature in the cylinder? Is the air circulation inside the cylinder adequate to      maintain a normal oxygen and carbon dioxide concentration? ... The article                   mentioned friction-fitted end caps. Does this mean that the patient was completely enclosed so they had no exposure to the outside at all?

We have updated the text with the following: [Lines 136-144] "The cylinder is mounted on a wooden cradle, and participants are supported (lying supine) on a retractable wooden bed, on a non-metal rail system which allows the participant to be inserted head-first until the helmet is approximately 1.65m into the cylinder. The shield end at the participant’s feet always remains open (i.e., there is an end cap at the participant’s head end only). Once inserted into the shield, the participant’s head is situated about 60cm from the end cap. A small bore in the end-cap (enabling external projection of visual stimuli; not used here) allows air flow through the cylinder (no heating or cooling is required). Ambient lighting from the surrounding room allows participants to see inside the cylinder. "

 

2.5.        Did the rigid helmet fit each patient equally well? Did it press or squeeze the head        for any patients?

We have added the following text: [Lines 181-184] "Layers of foam padding were used to cushion the back of the participant’s head against the helmet and, when necessary, additional padding was inserted to ensure that the participant was as comfortable as possible and to reduce head movement inside the helmet."

 

2.6.        Was it quiet in the cylinder, or were there any sources of noise (I don’t know if these    optical sensors are silent)?

We have added the following: [Lines 161-163] "The sensors are silent and operate at a maximum temperature of ~30°C, meaning they can be placed safely and comfortably against the head."

 

2.7.        How long were patients typically inside the cylinder?

We have added the following: [Lines 191-192] "Each participant was inside the scanner for approximately 30 minutes".

 

2.8.        Did any complain of headaches, anxiety, or claustrophobia afterwards?

No. We have added the following text: [120-123] "During recruitment and informed consent, potential participants were advised that individuals with claustrophobia might experience anxiety inside the shield, and that they should consider this before agreeing to participate." And [195-196] "After exiting the scanner, each participant was asked about their experience of the experiment. None reported excessive discomfort."

 

2.9.        Did any patient refuse to complete the study because they were not comfortable          inside the cylinder? ... Could the researchers communicate with the patient during             the experiment?

No participants withdrew from the study. We have updated the text with the following: [Lines 192-194] "Between tasks, a member of the research team communicated verbally with the participant to inform them of the upcoming task, and to ensure that they were still comfortable inside the scanner."

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I am pleased to see the additional sections on HFC as well as my other suggestions, however I am not sure I can come to the same conclusions as the authors on the effectiveness of HFC on the raw data. SNR is one measure, but I am concerned that the SEF responses are being attenuated by HFC. I think it would be best to offer some sort of simulation evidence using the lead fields to show that it is not removing actual brain signal from the data. The SNR ranges at sensor level when HFC is used seem to be quite large in comparison - is there one particular driver of this, because the mean and median values are quite similar? 

It is also difficult to compare the with and without HFC results in the figures, I think the supplementary figure should show both for ease of interpretation. 

Author Response

Reviewer comment #1:

I am pleased to see the additional sections on HFC as well as my other suggestions, however I am not sure I can come to the same conclusions as the authors on the effectiveness of HFC on the raw data. SNR is one measure, but I am concerned that the SEF responses are being attenuated by HFC. I think it would be best to offer some sort of simulation evidence using the lead fields to show that it is not removing actual brain signal from the data. The SNR ranges at sensor level when HFC is used seem to be quite large in comparison - is there one particular driver of this, because the mean and median values are quite similar?

Response:

It seems that our choice of wording in the revised manuscript led to ambiguity about our interpretation of the HFC results. To be clear: we agree with the reviewer that HFC subtracted much of the true signal from the recording via lead field attenuation. Indeed, the simulation work in the original HFC paper indicated that HFC will attenuate 1-2 dB (or 20-35%) of the true signal when applied to whole-head single-axis sensor arrays (Tierney et al., 2021). This is in line with our findings, which show ~50% lead field attenuation with a partial coverage array, as we would expect higher attenuation with fewer sensors.

With lead field attenuation established in our data, we recalculate SNR. We find that HFC was marginally helpful at the sensor level because it subtracted relatively more noise than true signal, leading to improved SNR.

We have tried to make this position more explicit on lines 646-662 (see text below with highlighted additions):

At the sensor level, applying HFC dramatically reduced the magnitude of SEF responses and movement-related mu/beta modulation. This indicates that HFC subtracted a large proportion of brain signal from our data via lead field attenuation. This finding is consistent with previous work showing that HFC can cause substantial (~20-35%) lead field attenuation when applied to simulated single-axis sensors in a full-coverage array (Tierney et al., 2021). In our data, this reduction at the sensor level was closer to 50%, reflecting the diminished effectiveness of HFC in partial-coverage arrays. These effects were less pronounced at the source level, likely due to the effectiveness of source estimation at noise reduction. In source space, we saw a slight reduction in SEF magnitude across all ROIs, while oscillatory (mu/beta) power change from baseline appeared unaffected.

Despite the clear attenuation of neural responses induced by HFC (particularly at the sensor level), we found that performing analyses without HFC generally reduced SNR at both the sensor and source levels. For example, skipping HFC reduced the maximum SNR across ROIs by ~65% (median across subjects), and also shifted the location (channel/ROI) at which SNR was maximal towards posterior areas - away from central areas where the SEF is typically observed. These results suggest that, although HFC lead field attenuation removes a large component of signal generated by the brain, it was still marginally beneficial in that it removed relatively more noise from the recordings than true signal.

We have also modified lines 760-763:

Moreover, modern preprocessing methods such as HFC, previously applied to data from systems with whole-head coverage, may improve SNR in partial-coverage systems such as this one. However, this must be assessed on a case-by-case basis dependent on the sensor positions and orientations and the signal(s) of interest. In general, researchers can expect higher HFC lead field attenuation (i.e., lost brain signal) with partial coverage, as compared to a full-coverage system.

We have not conducted the suggested simulation analysis because simulation data already exists showing lead-field attenuation with HFC (Tierney et al., 2021). Given this, simulation analysis would almost certainly show that HFC causes lead field attenuation, which is evident in our empirical results. Extending Tierney’s simulations to partial coverage is interesting, but the likely result is obvious (even more attenuation) and we feel that these simulations are beyond the scope of our paper. A fulsome simulation study would simulate many different sensor coverage schemes for many different numbers of sensors and various regions of interest – it is its own paper.  Further, by our reading of the reviewer's feedback, this suggestion appears predicated on the notion that HFC might not cause true signal loss (which is not our position).

Reviewer comment #2:

It is also difficult to compare the with and without HFC results in the figures, I think the supplementary figure should show both for ease of interpretation. 

Response: We have updated the Supplementary figure to include both sets of results.

Author Response File: Author Response.pdf