1

PEER REVIEW HISTORY

BMJ Open publishes all reviews undertaken for accepted manuscripts. Reviewers are asked to

complete a checklist review form (http://bmjopen.bmj.com/site/about/resources/checklist.pdf) and

are provided with free text boxes to elaborate on their assessment. These free text comments are

reproduced below.

ARTICLE DETAILS

TITLE (PROVISIONAL) Using Structural and functional MRI as a Neuroimaging Technique

to Investigate Chronic Fatigue Syndrome/ Myalgic

Encephalopathy: A Systematic Review

AUTHORS Almutairi, Basim; Langley, Christelle; Crawley, Esther; Thai, Ngoc Jade

VERSION 1 – REVIEW

REVIEWER Dr Gordon Waiter University of Aberdeen UK

REVIEW RETURNED 13-Jun-2019

GENERAL COMMENTS This systematic review seeks to identify the applicability of MRI based neuroimaging to assess chronic fatigue syndrome/myalgic encephalopathy. The systematic review was conducted following the appropriate methodology with a clear method to determine bias. The variability of the results in the papers included show that this review is timely and will help focus research in this topic. I recommend publication.

REVIEWER Nils Berginström Department of Psychology/Department of Community Medicine and Rehabilitation, Umeå University, Swede

REVIEW RETURNED 17-Jun-2019

GENERAL COMMENTS The authors present a systematic review of both structural and functional MRI findings in patients with Chronic Fatigue Syndrome/Myalgic Encephalomyelitis (CFS/ME). This study is the first to systematically review such studies, and is thus contributing to the field The article is mainly well written and the results are well discussed, but there are room for improvement. Abstract Good summary of the study, but do need some proof reading (e.g. line 4 in results). Introduction Balanced and good summary of the field. Methods Line 31: I would suggest spelling out all specific abbreviations used. The quality and risk of bias assessment description I methods is not coherent with the tables, unless I completely misunderstand it.

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

2

For example in the fMRI table (4) all studies report all of the criteria (Y), but all are low quality. How is this possible? This should be clarified or altered in the text or in tables. Results The headlining is quite confusing. 4. Is structural imaging, while 4.1 (a subheadline to structural imaging) is functional MRI findings. Please change this to avoid confusion. 4.1.2 Radiological Reporting. This section should be clarified here, in methods or under headline 3.2 Image Analysis sMRI. How did these visual inspections occur? Were patients compared to healthy controls? For example, white matter hyperintensities are quite common in normal aging, was this accounted for? And on a common note under this section, what does “abnormalities” (line 11) stand for? 4.1.3 White matter & Gray Matter volume. Line 35: “Changes in white matter observed onT2-weighted images in right middle temporal lobe were related to cognition demonstrated that white matter volume is negatively correlated with CFS/ME disease duration”. This sentence is not understandable. Should it be cut into to two or more sentences? And is should also be proof read, because there are some words and spaces missing. Line 44: “Grey matter volume reduction has been associated with functional deficiencies, and that may be influenced by pain, illness or age factor, thereby forcing the participants with CFS/ME to suffer from severe fatigue” This is interesting, and should be a part of the discussion section. However, I do not understand how pain, illness or age, “forces” patients to become fatigued. This should be clarified in the discussion section mentioned above. 4.2.2 Resting-state fMRI & Functional connectivity. Line 9: Funny sentence: “All four of the five”, should be altered. 4.2.3 fMRI & Cognition – Memory. In line 42 it is stated that: “in the complex and more challenging conditions (2-and 3-back conditions), participants with CFS/ME showed decreased activation in dorsolateral prefrontal and parietal cortices”. Similar statements can be seen under headlines reward and motivation, sensory information processing tasks and emotional conflict. This is different from the statement in the abstract (mainly increased activation). Increased/Decreased activation in clinical groups as compared to healthy controls is often dependent on task complexity or duration (See for example Bryer et al, 2013 Journal of the International Neuropsychological Society for a rewiew in patients with TBI.) A discussion on why some studies indicate increased or decreased activation is needed in the discussion section. Discussion In addition to what has been mentioned above, limitations of the present study should also be discussed.

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

3

Tables Please include your reference numbers in the tables, so these studies are more easily found in your reference list.

REVIEWER steve Quinn Swinburne University of Technology

REVIEW RETURNED 04-Nov-2019

GENERAL COMMENTS This systematic review assesses the impact and contribution of sMRI and fMRI on better understanding CFS/ME. 35 studies since 1991 informed this systematic review. This is a well-written paper on an important topic. The methodology is sound as far as it goes. The research question “evaluate sMRI and fMRI studies in CFS/ME” is quite broad, and the authors indicate that this is the first systematic review of neuroimaging studies that have investigated CFS/ME using MRI. For sMRI the authors talk about the reduction/increase in white/grey matter lesion. For fMRI decreased functional connectivity is mentioned. These attributes are quantitative and the report could be enhanced by including meta-analyses plus forest plots in this paper. This would add credibility to the claims in the first few sentence in the Conclusion in the Abstract. The results section need to contain more than descriptive statements about the number of studies, existing to date. Whilst this information is important, the authors need to synthesise the result of these studies to add to the body of knowledge. It is not sufficient to state in section 3.3 that Nichols’ criteria for risk of bias has been applied and it is contained in tables 3 and 4. The authors need to interpret this for the reader. The statements in the conclusion abstract appear to be already known as evidenced by the number of existing studies, e.g. “MRI have demonstrated the potential for significant insights into CFS/ME” or not proven in the text. For example, the authors note that the sample size of most studies is very small in their opinion and conclude that a large study would be helpful, but the real question is whether the sample sizes are adequate for the studies conducted. Heterogeneity, is mentioned several time in passing as a possible explanation for inconsistent findings without providing evidence of between-study differences to support this. There is no limitation section.

REVIEWER Elena A Shishkina Urals Research Center for Radiation Medicine, URCRM

REVIEW RETURNED 20-Dec-2019

GENERAL COMMENTS One of the main source of bias is the spatial resolution of image. This parameter is specifically important for combined analysis of sMRI data. Another one is the signal-to-noise ratio. That is quite important for reliability of fMRI. These two parameters should be indicated in the comparative analysis of the methods.

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

4

VERSION 1 – AUTHOR RESPONSE

Response to Reviewer 1:

Thank you for your time and consideration of this manuscript.

Response to Reviewer 2:

Thank you for your very comprehensive review of this paper. We have updated and edited the

paper to incorporate your feedback. We have responded to individual points as detailed in

italics after each section. You will also find our corrections written in blue where applicable in

the main manuscript, Tables, Figures or Supplementary Material.

1) Reviewer #1: Good summary of the study but do need some proof reading (e.g. line 4 in

results).

1) Response: Apologies for missing the full stop. We have altered our result section in the

abstract according to address the editor comments to the following: “[Structural MRI studies

report differences in CFS/ME brain anatomy in grey and white matter volume, ventricular

enlargement and hyper-intensities. Three studies report no neuroanatomical differences

between CFS/ME and healthy control. Task-based fMRI investigated working memory,

attention, reward and motivation, sensory information processing and emotional conflict. The

most consistent finding was CFS/ME produced increased activations and recruited additional

brain regions. Tasks with increasing load or complexity produced decreased activation in task

specific brain regions.]”.

2) Methods: Line 31: I would suggest spelling out all specific abbreviations used.

2) Response: All abbreviations have been added to the methods as the following (highlighted

in blue): “[ We used the following key words (and abbreviations) for CFS/ME: “chronic fatigue

syndrome”, “fatigue syndrome, chronic”, “myalgic encephalomyelitis”, “myalgic

encephalopathy”, “CFS”, “ME” or “CFS/ME”. To detect all structural and functional studies

which used magnetic resonance imaging in participants with CFS/ME, we used the following

key words for imaging techniques: “magnetic resonance imaging”, “MRI”, “structural MRI”,

“sMRI”, “functional magnetic resonance imaging”, “functional MRI”, “fMRI”, “resting state

functional magnetic resonance imaging”, “resting-state functional magnetic resonance

imaging”, “resting-state functional MRI”, “resting state functional MRI”, “rsfMRI” and “rsfMRI”.]”. The

full search strategy is provided in supplementary information.

3) The quality and risk of bias assessment description I methods is not coherent with the

tables, unless I completely misunderstand it. For example, in the fMRI table (4) all studies

report all of the criteria (Y), but all are low quality. How is this possible? This should be clarified

or altered in the text or in tables.

3) Response: Our quality assessment criteria are based on minimisation of bias. The tables

were made to show the risk of bias levels which indicates the quality of those studies. For

tables 3 and 4, the last column indicates low risk not low quality. We have moved the quality

assessment outcomes to section 4 and amended the text as follows for clarity:

“[4. Quality assessment for risk of bias

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

5

We applied the criteria from Nichols et al. (2017) to assess study quality for risk of bias [4].

See table 3 and 4 for risk of bias assessment. The sMRI studies risk of bias assessment

showed that study quality was highly variable. All fMRI studies were assessed to have low risk

of bias and therefore considered high quality.]”

We have also amended the table legend to make this clearer:

“[ Table 3: Shows risk of bias assessment for structural MRI studies.]”

“[Table 4: Shows risk of bias assessment for functional MRI studies.]”.

4) Results: The headlining is quite confusing. 4. Is structural imaging, while 4.1 (a subheadline

to structural imaging) is functional MRI findings. Please change this to avoid confusion.

4) Response: It has been corrected. Now the structural results are under the results section:

“[ 3.3. Structural MRI Results]”.

5) 4.1.2 Radiological Reporting. This section should be clarified here, in methods or under

headline 3.2 Image Analysis sMRI. How did these visual inspections occur?

5) Response: We have added a couple of sentences to explain that neuroradiologists have

been asked to examine and report the images of CFS/ME and HC: “[To evaluate and compare

the sMRI of CFS/ME and HC, these studies used one reviewer [19], two neuroradiologists [17,

18, 21, 22] or three neuroradiologists [20] to visually inspect the images. Two MRI studies

found ventricular enlargement [17, 22] While three studies reported white matter hyperintensities or

abnormalities which were defined as lesions identified by high signal intensity on

T2 or proton density-weighted pulse sequences [17, 20, 21].]”.

6) Were patients compared to healthy controls? For example, white matter hyperintensities

are quite common in normal aging, was this accounted for?

6) Response: all studies considered age and sex match as a solution to account for age and

sex factors. Therefore, age has been accounted for in their opinion. In the visual inspection

ones, CFS/ME brain images were compared to HC brain images. In term of automated

analysis methods, 10/13 took into account the age factor in their analysis. Therefore, the

following sentences has been added at the beginning of the paragraph: “[To evaluate and

compare the sMRI of CFS/ME and HC, these studies used one reviewer [19], two

neuroradiologists [17, 18, 21, 22] or three neuroradiologists [20] to visually inspect the images.

Two MRI studies found ventricular enlargement [17, 22] While three studies reported white

matter hyper-intensities or abnormalities which were defined as lesions identified by high

signal intensity on T2 or proton density-weighted pulse sequences [17, 20, 21]. In these

studies, age was only accounted for by using age-matched HC.]”.

We added the following sentences to the future direction section after these sentences to highlight the

importance of age when looking at neuroanatomical differences:

“An additional important aspect is the use of longitudinal MRI data. This is done to compare CFS/ME

results at two or more time points and find differences in the images obtained with the help of

automated analysis methods but keeping in mind the effect of aging [7, 86, 87] when looking purely

for structural changes: “[ In this systematic review, out of the thirteen quantitative studies, 11

corrected for age in their statistical analysis.]”.

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

6

Also, we had a sentence in the future direction to emphasize the importance of age when comparing

between studies of paediatric and adult CFS/ME populations:

“[Brain white matter volume increases linearly with age in adolescence [88] and given the prevalence

of CFS/ME in this age group [89], more research is required to determine if there are distinct

neurobiological markers comparable to studies in adults with CFS/ME.]”.

7) On a common note under this section, what does “abnormalities” (line 11) stand for?

7) Response: It stands for lesions as defined by high signal intensity on T2 or proton densityweighted

pulse sequences. We added the following sentence to clarify it: “[Two MRI studies found ventricular

enlargement [17, 22] While three studies reported white matter hyperintensities or abnormalities

which were defined as lesions identified by high signal intensity on T2 or proton density-weighted

pulse sequences [17, 20, 21].]”.

8) 4.1.3 White matter & Grey Matter volume: Line 35: “Changes in white matter observed onT2-

weighted images in right middle temporal lobe were related to cognition demonstrated that white

matter volume is negatively correlated with CFS/ME disease duration”. This sentence is not

understandable. Should it be cut into to two or more sentences? And is should also be proofread,

because there are some words and spaces missing.

8) Response: The sentences have been edited accordingly and proofread:

“[ Authors were able to demonstrate that white matter volume is negatively correlated with CFS/ME

disease duration [5, 7]. This means that white matter volumes decreased with increase disease

duration [5, 7].]”

9) Line 44: “Grey matter volume reduction has been associated with functional deficiencies, and that

may be influenced by pain, illness or age factor, thereby forcing the participants with CFS/ME to suffer

from severe fatigue” This is interesting, and should be a part of the discussion section. However, I do

not understand how pain, illness or age, “forces” patients to become fatigued. This should be clarified

in the discussion section mentioned above.

9) Response: The word forcing was not accurate. Therefore, the sentence has been changed to:

“[Grey matter volume reduction has been associated with functional deficiencies, that may be

influenced by pain [38, 39] illness or age factors, thereby having a detrimental impact on quality of life

for participants with CFS/ME [11, 14].]”.

We have added to structural MRI discussion the relevance of pain being a symptom in many other

chronic diseases that has been associated with grey matter volume changes in sMRI and therefore

not unique to CFS/ME:

“[Pain is an important factor which happens at any site from the cerebral cortex to the spinal cord and

believed to be caused by maladaptive functional or structural plasticity of the nociceptive system [50].

Pain is a common symptom in CFS/ME but not a primary symptom for diagnosis. Altered brain

morphology on sMRI has been reported in many types of pain disorders, including chronic back pain

[51-53], chronic tension-type headache [54], fibromyalgia [55, 56], migraine [57-59], and somatoform

pain disorder [60]. Thus, not unique to CFS/ME.]”.

10) 4.2.2 Resting-state fMRI & Functional connectivity. Line 9: Funny sentence: “All four of the five”,

should be altered.

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

7

10) Response: The sentence has been changes to: “[Four out of the five resting state fMRI studies

reported decreased functional connectivity in participants with CFS/ME compared to healthy controls

[23, 26, 35, 42].]”.

11) 4.2.3 fMRI & Cognition – Memory. In line 42 it is stated that: “in the complex and more challenging

conditions (2-and 3-back conditions), participants with CFS/ME showed decreased activation in

dorsolateral prefrontal and parietal cortices”. Similar statements can be seen under headlines reward

and motivation, sensory information processing tasks and emotional conflict. This is different from the

statement in the abstract (mainly increased activation). Increased/Decreased activation in clinical

groups as compared to healthy controls is often dependent on task complexity or duration (See for

example Bryer et al, 2013 Journal of the International Neuropsychological Society for a review in

patients with TBI.)

A discussion on why some studies indicate increased or decreased activation is needed in the

discussion section.

11) We have changed the line in abstract to:

“[The most consistent finding regardless of task was CFS/ME often produced increased activation

patterns and recruited additional brain regions. compared to healthy controls. Tasks with increasing

load or complexity produced decreased activation in task specific brain regions. Resting state fMRI

mostly reported decreased functional connectivity in CFS/ME.]”.

We have discussed the implications of the increases and decreases of activation patterns reported by

fMRI studies in the fMRI discussion section, we have added additional information to this section and

highlighted in blue:

“[Task differences in difficulty might play a major role in why some studies reported increase

activation while others reported decreased activation. Bryer et al 2013 meta-analysis of fMRI studies

of memory function in Traumatic Brain injury patients concluded that the primary reason for

discrepancy in activation patterns across studies is attributable to task classification, hyperactivation

be associated with continuous memory tasks and hypoactivation being more prominent in discrete

memory tasks [70]. There have been a wide variety of tasks used in fMRI studies to assess

differences between participants with CFS/ME and healthy controls. When resting-state fMRI or

simple tasks, were employed, participants with CFS/ME showed decreased functional connectivity in

various brain regions [23-26, 28, 35, 40, 42, 44]. However, when more challenging tasks are

employed, participants with CFS/ME exhibit widespread increased activation in task related regions

when compared to healthy controls [29, 31, 33, 34, 43, 44]. Most of the CFS/ME participants

performed at a similar level to healthy controls and it is not clear whether the increased activation was

due to the increase in the task difficulty or because CFS/ME participants are trying harder. This

widespread activation may lead to an increase in demand on neural resources such as oxygen and

glucose which in turn would lead to fatigue [10]. Fatigue and lower performance have been

associated with brain activity increase while performing a high-effort cognitive task [9, 27, 29]. It has

been hypothesized that severe fatigue consumes a significant amount of attentional resources in term

of recruiting additional brain regions for cognitive compensation to perform better in dual task

depending on the degree of mental effort [29, 33]. Caseras et al. (2006) suggested that the fear of

being fatigued leads the CFS/ME group to avoid activity [33]. Impaired reward processing was

suggested to decrease motivation to learn in adolescents with CFS/ME [28], while participants

inability to engage the part of the brain (left amygdala and left midposterior insula) that responds to

conflict suggested an abnormal salience network functioning in term of affect and cognition [25]. The

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

8

increased activity in task related areas was hypothesized to be a result of cognition and emotion

deficits in participants with CFS/ME [24, 33] and impaired reward processing in adolescents [29].

Reduction in activation was interpreted as when the task difficulty increases, participants with

CFS/ME failed to recruit working memory regions to the same level as the healthy controls [33].

“[The heterogeneity of tasks, behaviours and cognitive processes across the fMRI studies makes it

difficult to discern how much of the increase or decrease activations reported in relation to task

difficulty or demand is associated with increasing cognitive fatigue. Understanding the impact of

fatigue on brain function will be critical to our understanding of CFS/ME.]”.

Discussion

12. In addition to what has been mentioned above, limitations of the present study should also be

discussed.

12. We have added a section 6: limitations and highlighted the main limitation with our systematic

review

“[The main limitation of the present systematic review is that there was insufficient data for meta-

analysis. Meta- analysis of neuroimaging data can take 2 approaches, image-based or coordinate

based analysis. Image-based analysis requires the statistical images of the data and this is not often

available due to data sharing issues e.g. data protection and other restrictions. Therefore, most

neuroimaging meta-analysis is co-ordinated based as these are reported in the published research

and because spatial normalization of images into standardized coordinates as anatomical addresses

within a reference space being applied to human neuroimaging data for decades [77, 78]. In order to

perform an appropriate coordinate based meta-analysis some minimum criteria need to be met.

Firstly, power of the meta-analysis. For co-ordinate based meta-analysis the activation likelihood

estimation (ALE) method is conventionally applied [79], or a revised ALE algorithm [80]. Based on a

recent simulation study performed by Eickhoff et al (2016), a recommendation was made to include at

least 17–20 experiments in ALE meta-analyses for sufficient power to detect smaller effects and

ensure results are not driven by single experiments [81]. For sMRI we have 19 studies, 6 of these are

visual inspection thus subjective reporting without quantitative coordinate data. Of the 13 quantitative

studies only 6 studies reported coordinates. A meta-analysis on 6 sMRI studies would be severely

under powered.

For fMRI we have 16 studies in total, 5 fMRI studies are rs-fMRI and 11 task-based fMRI. The

methodological differences between them precludes us from combining the results to perform a

coordinate based meta-analysis, they did not all meet the two minimum criteria. Firstly, all studies

must be whole brain analysis and secondly, they must use the same standardized coordinates

system. Four rs-fMRI use ROI seed regions and 1 used whole brain approach for calculating

connectivity. For task-based fMRI studies the task selection criteria are critical. In our systematic

review we identified 11 task-based fMRI studies. However only 2 studies use the same PASAT task.

The remaining 9 studies all use different tasks and more importantly each task is designed to examine

a different cognitive, sensory or physical function. Therefore, the heterogeneity of task-based fMRI

studies prohibits the creation of task selection criteria for meta-analysis.]”

And Future directions we have added the following

“[Finally, we have limited this systematic review to include only neuroimaging studies that have used

Structural or functional MRI methods. However other neuroimaging techniques have been used to

investigate CFS/ME these include Single-photon emission computerized tomography (SPECT),

electroencephalogram (EEG) and Magnetic Resonance Spectroscopy and Diffusion Tensor Imaging

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

9

(DTI). These methods measure different neurophysiology from sMRI and fMRI. Future studies using

multi-model imaging approaches could overcome some of limitations of a single method alone.]”

13. Tables

Please include your reference numbers in the tables, so these studies are more easily found in your

reference list.

13. We have added the reference numbers in all the tables, and these are highlighted in blue in the

revised manuscript

Response to Reviewer 3:

1) For sMRI the authors talk about the reduction/increase in white/grey matter lesion. For fMRI

decreased functional connectivity is mentioned. These attributes are quantitative, and the report could

be enhanced by including meta-analyses plus forest plots in this paper. This would add credibility to

the claims in the first few sentences in the Conclusion in the Abstract.

1) Response: the following paragraph has been added to the main manuscript in limitations section 6:

“[The main limitation of the present systematic review is that there was insufficient data for meta-

analysis. Meta- analysis of neuroimaging data can take 2 approaches, image-based or coordinate

based analysis. Image-based analysis requires the statistical images of the data and this is not often

available due to data sharing issues e.g. data protection and other restrictions. Therefore, most

neuroimaging meta-analysis is co-ordinated based as these are reported in the published research

and because spatial normalization of images into standardized coordinates as anatomical addresses

within a reference space being applied to human neuroimaging data for decades [77, 78]. In order to

perform an appropriate coordinate based meta-analysis some minimum criteria need to be met.

Firstly, power of the meta-analysis. For co-ordinate based meta-analysis the activation likelihood

estimation (ALE) method is conventionally applied [79], or a revised ALE algorithm [80]. Based on a

recent simulation study performed by Eickhoff et al (2016), a recommendation was made to include at

least 17–20 experiments in ALE meta-analyses for sufficient power to detect smaller effects and

ensure results are not driven by single experiments [81]. For sMRI we have 19 studies, 6 of these are

visual inspection thus subjective reporting without quantitative coordinate data. Of the 13 quantitative

studies only 6 studies reported coordinates. A meta-analysis on 6 sMRI studies would be severely

under powered.

For fMRI we have 16 studies in total, 5 fMRI studies are rs-fMRI and 11 task-based fMRI. The

methodological differences between them precludes us from combining the results to perform a

coordinate based meta-analysis, they did not all meet the two minimum criteria. Firstly, all studies

must be whole brain analysis and secondly, they must use the same standardized coordinates

system. Four rs-fMRI use ROI seed regions and 1 used whole brain approach for calculating

connectivity. For task-based fMRI studies the task selection criteria are critical. In our systematic

review we identified 11 task-based fMRI studies. However only 2 studies use the same PASAT task.

The remaining 9 studies all use different tasks and more importantly each task is designed to examine

a different cognitive, sensory or physical function. Therefore, the heterogeneity of task-based fMRI

studies prohibits the creation of task selection criteria for meta-analysis.]”.

We have also changed our conclusion section in the abstract to address the editor’s comments: In the

Conclusions in the Abstract, please be clearer about your overall findings, reach some sort of

conclusion, state the limitations of neuroimaging and suggest what the implications are going forward.

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

10

“[There was insufficient data to define a unique neural profile or biomarker of CFS/ME. This may be

due to inconsistences in finding neuroanatomical differences in CFS/ME and the variety of different

tasks employed by fMRI studies. But there are also limitations with neuroimaging. All brain region

specific volumetric differences in CFS/ME were derived from voxel-based statistics that are biased

toward group differences that are highly localized in space. fMRI studies demonstrated increases and

decreases in activations patterns in CFS/ME, this may be related to task demands. However, fMRI

signal cannot differentiate between neural excitation and inhibition or function-specific neural

processing. Many studies have small sample sizes and did not control for the heterogeneity of this

clinical population. We suggest that with robust study design, subgrouping and larger sample sizes

future neuroimaging studies could potentially lead to a breakthrough in our understanding of this

illness.]”

2) The results section needs to contain more than descriptive statements about the number of

studies, existing to date. Whilst this information is important, the authors need to synthesise the result

of these studies to add to the body of knowledge.

2) Response: Thanks for your comment. It has been addressed by putting the synthesise of studies

results that was previously section 4 under Section 3 Results and changing the headings from MRI

findings to MRI results and fMRI findings has been changed to fMRI results.

3) It is not sufficient to state in section 3.3 that Nichols’ criteria for risk of bias has been applied and it

is contained in tables 3 and 4. The authors need to interpret this for the reader.

3) Response: The use of Nichols et al. (2017) is to show that these criteria are important and useful in

order to replicate those studies. We have added the following in the manuscript: “[These criteria have

been set with the aim for reproducible research for neuroimaging studies using MRI. Stating clearly

the recruitment procedure, Inclusion/exclusion criteria, population demographics and comparison

group enables a critical reader to evaluate and decide whether this sample is bias or can be

generalised to other populations [4].]”.

4) The statements in the conclusion abstract appear to be already known as evidenced by the number

of existing studies, e.g. “MRI have demonstrated the potential for significant insights into CFS/ME” or

not proven in the text.

For example, the authors note that the sample size of most studies is very small in their opinion and

conclude that a large study would be helpful, but the real question is whether the sample sizes are

adequate for the studies conducted.

4) Response: we have added to section 6 more details on neuroimaging studies sample size to

address and qualify our statements on small sample sizes:

“[Our systematic review has highlighted a limitation of the fMRI studies in CFS/ME is the small sample

sizes. Empirical and simulation studies conducted by Desmond & Glover 2002 [71] found that to

achieve 80% power at the single voxel level for typical activations in fMRI studies with thresholds

correcting for multiple comparisons a sample size of 24 is required. We found that 15 of the 16 fMRI

studies had a patient sample size of less than 24. Studies with low power reduce the likelihood of

detecting a true effect and increases the likelihood of false positive. However, this is not unique to

neuroimaging studies in CFS/ME, fMRI studies have been criticized for being underpowered due to

small sample sizes resulting in overestimates of effect size and low reproducibility [72, 73].]”.

5) Heterogeneity is mentioned several times in passing as a possible explanation for inconsistent

findings without providing evidence of between-study differences to support this.

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

11

5) Response: Our main point about Heterogeneity in this systematic review has been about the

patient population under study and we have emphasized this further in our revisions in following

section 5.1:

“[CFS/ME does not currently have any biomarkers or clinical signs therefore diagnosis is based upon

self-reported symptoms and excluding alternative explanations for diagnoses. The use of self-

reported symptoms leads to doubt about the validity of CFS/ME as an aetiologically homogeneous

diagnosis [62, 63]. This has in turn produced research to empirically define cases and subgroups

examining the heterogeneity of CFS/ME. Hickie et al 1995 used symptoms and demographics to

empirically define a core group and a smaller polysymptomatic subgroup [64]. A more recent study by

Williams et al 2017 used Latent class analysis to empirically define subgroups in a sample of 541

CFS/ME patients and found 5 subgroups [65].]”.

And

“[Shan et al. (2016) showed a reduction in left inferior fronto-occipital fasciculus (white matter) when

doing a six years longitudinal study [17] but this was not consistent with Perrin et al. (2010) who

investigated this after a year [20]. These differences in findings could be due to differences in timing.

CFS/ME might be a slow progressing illness and neuro dysfunction is related to length of illness [5, 7].

Alternatively, it could be due to different populations studied, as none of the studies defined their

CFS/ME population accounting for subgroups.]”

However, we have also highlighted the Heterogeneity between the studies and listed what these

difference are, for example section 5.1 “[Differences in methodology include using visual inspection,

computational analysis, different sample sizes and CFS/ME patients with different length of illness or

symptom severity.]”

6) There is no limitation section.

6) Response: Thanks for your comment. It has been addressed and we have included this limitation

section:

“[6. Limitations Our systematic review has highlighted a limitation of the fMRI studies in CFS/ME is the

small sample sizes. Empirical and simulation studies conducted by Desmond & Glover 2002 [71]

found that to achieve 80% power at the single voxel level for typical activations in fMRI studies with

thresholds correcting for multiple comparisons a sample size of 24 is required. We found that 15 of

the 16 fMRI studies had a patient sample size of less than 24. Studies with low power reduce the

likelihood of detecting a true effect and increases the likelihood of false positive. However, this is not

unique to neuroimaging studies in CFS/ME, fMRI studies have been criticized for being underpowered

due to small sample sizes resulting in overestimates of effect size and low reproducibility [72, 73].

All the studies in this systematic review did not report the signal to noise (SNR) in their MRI methods.

The SNR compares the level of the signal of interest to the level of background noise. In MRI studies

SNR is important for comparison between different MRI scanners, imaging protocols and MR

sequences [74]. Thus, limiting this systematic reviews’ ability to do a comprehensive comparative

analysis

The automated computational methods for investigating structural anatomical differences may be

superior to subjective visual inspection but does have some limitations. Voxel-based morphometric

analysis has been criticized for being significantly biased toward group differences that are highly

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

12

localized in space and of a linear nature, and poor at detecting group differences that are spatially

complex and subtle [75]. The fMRI studies report both increases and decreases in activations

patterns in CFS/ME compared to controls, while this may be related to task demands, caution must

also be taken when interpreting these results, bearing in mind fMRI is an indirect measure of neural

activity. The fMRI signal is derived from the Blood Oxygen level dependant (BOLD) contrast

mechanism i.e. haemodynamics of the brain. Currently we cannot easily estimate the cerebral

metabolic rate of oxygen (CMRO2) from the BOLD. Furthermore, haemodynamic responses are

sensitive to the size of the activated population, and less likely to detect cortical regions in when

stimulus- or task-related perceptual or cognitive capacities have sparse neuronal representation. It is

also not fully understood how neuromodulation might contribute to the spatiotemporal resolution of the

fMRI signal [76]. In recent years there has been a shift placing a greater emphasis on neural networks

underlying behaviour and cognition. A functional connectivity approach considering the neural

network difference between patients and healthy populations may lead to better understanding of how

disease affects brain function.

The main limitation of the present systematic review is that there was insufficient data for meta-

analysis. Meta- analysis of neuroimaging data can take 2 approaches, image-based or coordinate

based analysis. Image-based analysis requires the statistical images of the data and this is not often

available due to data sharing issues e.g. data protection and other restrictions. Therefore, most

neuroimaging meta-analysis is co-ordinated based as these are reported in the published research

and because spatial normalization of images into standardized coordinates as anatomical addresses

within a reference space being applied to human neuroimaging data for decades [77, 78]. In order to

perform an appropriate coordinate based meta-analysis some minimum criteria need to be met.

Firstly, power of the meta-analysis. For co-ordinate based meta-analysis the activation likelihood

estimation (ALE) method is conventionally applied [79], or a revised ALE algorithm [80]. Based on a

recent simulation study performed by Eickhoff et al (2016), a recommendation was made to include at

least 17–20 experiments in ALE meta-analyses for sufficient power to detect smaller effects and

ensure results are not driven by single experiments [81]. For sMRI we have 19 studies, 6 of these are

visual inspection thus subjective reporting without quantitative coordinate data. Of the 13 quantitative

studies only 6 studies reported coordinates. A meta-analysis on 6 sMRI studies would be severely

under powered.

For fMRI we have 16 studies in total, 5 fMRI studies are rs-fMRI and 11 task-based fMRI. The

methodological differences between them precludes us from combining the results to perform a

coordinate based meta-analysis, they did not all meet the two minimum criteria. Firstly, all studies

must be whole brain analysis and secondly, they must use the same standardized coordinates

system. Four rs-fMRI use ROI seed regions and 1 used whole brain approach for calculating

connectivity. For task-based fMRI studies the task selection criteria are critical. In our systematic

review we identified 11 task-based fMRI studies. However only 2 studies use the same PASAT task.

The remaining 9 studies all use different tasks and more importantly each task is designed to examine

a different cognitive, sensory or physical function. Therefore, the heterogeneity of task-based fMRI

studies prohibits the creation of task selection criteria for meta-analysis.]”.

Response to Reviewer 4:

1) One of the main sources of bias is the spatial resolution of image. This parameter is specifically

important for combined analysis of sMRI data.

Another one is the signal-to-noise ratio. That is quite important for reliability of fMRI. These two

parameters should be indicated in the comparative analysis of the methods.

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

13

Response: We have added a limitations section and here we have addressed how spatial resolution

is accounted for by comparative analysis methods through: “[spatial normalization of images into

standardized coordinates as anatomical addresses within a reference space being applied to human

neuroimaging data for decade]”

we have also highlighted that all the studies in this systematic review did not report the signal to noise

(SNR) in MRI methods: “[All the studies in this systematic review did not report the signal to noise

(SNR) in their MRI methods. The SNR compares the level of the signal of interest to the level of

background noise. In MRI studies SNR is important for comparison between different MRI scanners,

imaging protocols and MR sequences [74]. Thus, limiting this systematic reviews’ ability to do a

comprehensive comparative analysis.]”

VERSION 2 – REVIEW

REVIEWER Nils Berginström Department of Psychology, Umeå University, Sweden

REVIEW RETURNED 12-Feb-2020

GENERAL COMMENTS The authors have made a substantial revision, which have improved the manuscript a largely taken care of my previous concerns. However, there are some minor issues that should be revised before the paper is acceptable for publication: Proofreading The entire manuscript should be proofread, since there are some spelling and grammatical errors tha affects readability. Heading 3.3.1 The authors write: “These included both grey matter volume and white matter volume reduction, ventricular enlargement, white matter hyper-intensities, abnormalities and cortical thickening.” Please explain what abnormalities stands for in this case. Tables: Spell out all abbreviations used in tables, either within the tables or in a footnote.

REVIEWER Steve Quinn Swinburne University of Technology Melbourne, Australia

REVIEW RETURNED 04-Feb-2020

GENERAL COMMENTS My concerns have been addressed. Two small comments arising from the changes are: a) Limitations section – “studies with low power reduce the likelihood of detecting the true effect and increase the likelihood of a false positive”. It’s the opposite and studies with low power increase the likelihood of a false negative. The authors have misunderstood the reference (72) that they quote in support of this. b) The first sentence in the conclusion is clumsy. Better to say, “there is no evidence to support the assertion that…”.

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

14

VERSION 2 – AUTHOR RESPONSE

Reviewer: 2

Please leave your comments for the authors below The authors have made a substantial revision,

which have improved the manuscript a largely taken care of my previous concerns. However, there

are some minor issues that should be revised before the paper is acceptable for publication:

Proofreading

1. The entire manuscript should be proofread, since there are some spelling and grammatical errors

that affects readability.

1. Two authors have proofread the manuscript and made changes and we have uploaded a track

changes marked copy of the manuscript

2. Heading 3.3.1 The authors write: “These included both grey matter volume and white matter

volume reduction, ventricular enlargement, white matter hyper-intensities, abnormalities and cortical

thickening.” Please explain what abnormalities stands for in this case.

2.We have changed the word abnormalities to lesions as the authors that reported abnormalities in

CFS/ME sMRI refer to abnormalities as lesions identified by high signal intensity on T2 or proton

density-weighted pulse sequences

“[These included both grey matter volume and white matter volume reduction, ventricular

enlargement, white matter hyper-intensities, lesions and cortical thickening.]”

3. Tables: Spell out all abbreviations used in tables, either within the tables or in a footnote.

3. We have added footnote to tables 1 and 2 and fully spelled out all abbreviations. We have only

used the abbreviation CFS and HC in table 3 and 4 and these have been spelled out in the

manuscript.

Table 1: Summary of 19 sMRI studies in CFS/ME

* Healthy controls ** Some studies provided average age and others provided a range. *** Not

mentioned CDC = Centre for Disease Control CFS = Chronic Fatigue Syndrome, HC = Healthy

controls, FSL =FMRIB Software Library, SPM = Statistical parametric mapping

Table 2: Summary of 16 fMRI studies in CFS/ME

* Healthy controls ** Some studies provided average age and others provided a range. CDC =Centre

for disease control, CFS = Chronic Fatigue Syndrome HC = Healthy Control, XBAM = Individual Brain

Activation Maps, AFNI = Analysis of Functional NeuroImages, SPM =Statistical parametric mapping,

PASAT =Paced Auditory Serial Addition Test

Reviewer: 3

Please leave your comments for the authors below

My concerns have been addressed. Two small comments arising from the changes are:

1a) Limitations section – “studies with low power reduce the likelihood of detecting the true effect and

increase the likelihood of a false positive”. It’s the opposite and studies with low power increase the

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from

15

likelihood of a false negative. The authors have misunderstood the reference (72) that they quote in

support of this.

1a Reference 72 is Button, et al Power failure: why small sample size undermines the reliability of

neuroscience. Nat Rev Neurosci, 2013. 14(5): p. 365-76. In personal correspondence with Dr

Katherine Button she confirmed that her publication was making the claim that “studies with low

power reduce the likelihood of detecting the true effect (i.e., increased risk of false negatives) and

increase the likelihood of a false positive (by lowering the PPV of the test - i.e., low power reduces the

Positive Predictive Value of the test, and thus increases the chances of a positive result being a false

positive, especially for more exploratory hypotheses with low pre-study odds)”.

Therefore, we have changed the sentence in the manuscript for clarification

“[Studies with low power reduce the likelihood of detecting a true effect, increases the risk of false

negatives and the likelihood of false positives by reducing the positive predictive value (PPV) of the

test [74].]”

1b) The first sentence in the conclusion is clumsy. Better to say, “there is no evidence to support the

assertion that…”.

1b We have changed the first sentence of the conclusion as suggested to the following

“[In conclusion, there is no evidence to support the assertion that findings from neuroimaging studies

have found any clear biomarkers of CSF/ME.]”

on Septem

ber 2, 2021 by guest. Protected by copyright.

http://bmjopen.bm

j.com/

BM

J Open: first published as 10.1136/bm

jopen-2019-031672 on 30 August 2020. D

ownloaded from