Cortical-striatal network functional connectivity markers in poststroke fatigue: a single-centre fMRI case-control study protocol

Nature of the cortical–striatal network (CSN) and functional connectivity (FC)

The brain is organised as a set of widely distributed and interconnected networks, among which the CSN plays an important role in learning, motivation, addiction and reward-guided behaviour.1 2 The CSN comprises the basal ganglia, frontal cortex and the underlying white matter.3–5 The temporal correlations between the resting blood oxygenation-dependent signals in different brain regions (measured by resting-state functional MRI (fMRI)) are thought to reflect important inter-relationships among structures with related functions, and CSN FC abnormalities have been reported in neurological diseases.6 7

FC abnormalities in poststroke fatigue (PSF)

Structural and functional abnormalities in the CSN are hypothesised to play a key role in the pathogenesis of neurological disease-associated fatigue.7 8 Indeed, structural abnormalities such as infarcts,9 10 cerebral microbleeds (CMBs)11 and white matter abnormalities12–14 have been reported in patients with neurological disease-associated fatigue. In a task-evoked fMRI study of fatigue in patients with multiple sclerosis, fatigue was associated with decreased activation of the cortical frontal regions and caudate.15 Increased FC was also observed in participants with experiment-induced fatigue,16 and there is evidence suggesting that abnormal frontal FC contributes to fatigue in traumatic brain injury.17

Local structural damage due to stroke can result in dysfunction in remote regions connected to the lesion, and FC abnormalities have been reported in stroke.18 Hence, there is increasing interest in using resting-state fMRI measures to predict subsequent recovery19 and response to treatment20 in this patient population. The major advantage of resting-state fMRI over task-evoked fMRI is the capacity of the former to examine many networks concurrently.21 In a published resting-state fMRI study on poststroke depression, FC at 10 days post stroke was associated with depressive symptoms at a 3-month follow-up in 24 stroke survivors.22

In healthy individuals, mental fatigue is related to decreased FC, especially in the middle frontal gyrus23 and right angular gyrus.24 In patients with multiple sclerosis, fatigue is associated with decreased FC in the precentral gyrus25 and a weaker cortical-to-subcortical FC (between the posterior parietal cortex and right caudate head).26

Some small-scale fMRI studies have suggested that PSF is related to focal FC changes. In a study of 40 patients who had a mild stroke, there was no difference in FC between the fatigue and non-fatigue groups.27 In contrast, in a study by Schaechter et al, 12 patients who had a middle cerebral artery stroke, loss in the structural connectivity and FC between bilateral frontal cortex regions were found to play a role in PSF.28 In an experimental study of 59 patients who had a stroke, an imbalance in interhemispheric inhibitory effects between primary motor regions explained subjective PSF.29 Finally, in a 6-week clinical trial involving 26 patients, fronto–striato–thalamic FC predicted modafinil response for PSF.30 To date, there has been no published large-scale fMRI study on PSF.

Possible pathomechanism of the CSN in PSF

The basal ganglia play an important role in learning, motivation, addiction and reward-guided behaviour.2 They have wide projections to the prefrontal cortex, thus forming the CSN. The CSN theory of fatigue proposed by Chaudhuri and Behan8 focuses on the underlying mechanisms of fatigue. Their theory posits that fatigue is related to a failure in the integration of the limbic input and motor functions within the basal ganglia. Fatigue is partly due to a loss of motivation in self-initiated tasks.8 According to modern neuropsychological theory, fatigue can result from inappropriate effort output and outcome valuation owing to CSN dysfunction.31 It is plausible that grey and white matter lesions/ischaemia in the CSN disrupt the brain mechanisms of motivation, leading to fatigue symptoms.

Other predictors of PSF risk

The literature on the relationship between lesion location and PSF reports conflicting results.32 Some authors have reported the number of acute infarcts9 and internal capsule, brain stem, cerebellum and basal ganglia infarcts9 10 and caudate lesions,33 in particular, to be associated with PSF, although others have been unable to replicate these findings.34 35 Similarly, white matter hyperintensities (WMHs) have not been shown to be associated with PSF.10 33 36 Recent studies have suggested that CMBs11 may play a role in the development of PSF. Although there was no previous diffusion tensor imaging (DTI) study on the white matter integrity in PSF, white matter abnormalities in the anterior thalamic tracts and corpus callosum are reportedly related to fatigue in multiple sclerosis.14 PSF is a complex disorder, and a host of non-imaging factors contribute to its development. Its putative risk factors comprise both demographic (younger age and female sex) and clinical (neurological deficits, physical functioning, anxiety, depressive symptoms and sleep disturbances) factors.10 33 37–39

Predictors of PSF symptom severity

Physical functioning, anxiety and depressive symptoms40 are possible correlates of fatigue severity, although the volume of acute infarcts was not related to fatigue severity in a sample of 45 patients who had an acute stroke.41 There are no published data on the effects of WMHs or white matter fractional anisotropy (FA) or FC on PSF symptom severity. In an MRI study of multiple sclerosis, fatigue scores were correlated with frontal white matter lesion volume,12 whereas in a DTI study of this patient population, fatigue scores were correlated with CSN FA.13 Finally, a task-evoked fMRI study of fatigue in multiple sclerosis found inverse correlations between fatigue severity and activation of the cortical and subcortical areas (r=−0.61 to −0.63).42 Lassalle-Lagadec et al
22 identified correlations between FC and depressive symptoms in stroke survivors (r=0.49–0.77).

Predictors of the PSF outcome

The rate of PSF remission at 12–18 months post stroke varies from 18% to 36%,10 34 43–45 and lower levels of fatigue and anxiety at baseline are reported to predict fatigue remission in non-stroke samples.46 47 No imaging predictors of non-remission in PSF have been identified thus far.10

Sample size estimation

Power will be computed using PASS 2005, and 738 patients will be screened for PSF. Using 23.4%9 as an estimate of PSF frequency, we expect to identify 173 (738×23.4%) patients with PSF. The control group will consist of 173 age-matched and sex-matched stroke survivors without PSF. The sample size ratio between the PSF cases and matched controls will be 1.0 (173/173).

For hypothesis (a), stroke patients with PSF have reduced CSN FC, the sample size of 173 for the PSF cases and controls will provide 80% power in identifying any FC risk factors, assuming a small effect size (Cohen’s d=0.31). On recent study suggested neuroimaging marker, such as location of infarcts, had a small to medium effect on the risk of PSF.48

For hypothesis (b), reduced CSN FC is associated with more severe PSF, a sample of 173 patients with PSF will also provide 80% power in detecting a correlation coefficient of at least 0.21 between FC and Fatigue Severity Scale (FSS) scores. On recent study found the correlation coefficient between FC of certain brain networks and PSF severity was between of 0.4 and 0.5.27

For hypothesis (c), reduced CSN FC predicts the non-remission of PSF. Assuming a dropout rate of 20%, we expect that 138 (173×80%) patients with PSF will be assessed at the follow-ups. With a remission rate of 36%,34 this sample size will give rise to 80% power in identifying any FC predictors, assuming a medium effect size (Cohen’s d=0.57). We found certain predictors of PSF remission had medium-to-large effect size of remission of PSF.49

Recruitment of participants

This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology Statement research checklist.50 The planned study will be a prospective nested case–control study of stroke survivors. Details of recruitment are shown in figure 1. Patients will be recruited from the Acute Stroke Unit (ASU) of the Prince of Wales Hospital in Hong Kong. The ASU treats approximately 93% of all patients who had an acute stroke and were admitted to the hospital, with the majority of the remaining 7% sent to the neurosurgery unit. All patients who had an acute stroke consecutively admitted to the ASU over a period of 2 years (approximately 2 000) will be invited to participate in the study. A research assistant will visit the ASU daily to identify all eligible patients and obtain their written consent.

Figure 1

Flowchart of the study.

It is estimated that approximately 80% of these 2000 patients (n=2000×80%=1600) will have experienced an ischaemic stroke.51 MRI examination is expected to be contraindicated in 10% of patients, leaving 1440 (1600×90%) eligible patients. Past experience suggests that the dropout rate will be around 36%.52 The major reasons for dropping out in our previous study were death (35% of all dropouts), dialect barrier (19%), refusal to participate (14%), prolonged (more than 3 months) hospitalisation (11%) and physical frailty (8%).52 Hence, the number of potential participants is estimated to be 922 (1440×(100%−36%)). Finally, 20% of these 922 participants are expected to be excluded due to dementia (10%), aphasia (7%) and/or history of depression (35%), leaving 738 eligible participants (922×(100%−20%)).

Eligibility criteria

Data collection

Details of the data collection schedule are shown in table 1. The number of patients excluded and the reasons for their exclusion will be recorded. Data on age, sex, level of education, vascular risk factors (eg, smoking, hypertension, diabetes mellitus and hyperlipidemia) and date of stroke onset will be collected for all participants. Data on neurological impairment in terms of the NIHSS total score at admission will be extracted from the stroke registry, which will be maintained by a full-time, well-trained research nurse who is also responsible for the administration of the NIHSS.

Assessment of PSF at baseline

The principal investigator will conduct interviews at 3 months post stroke (P1) at a research clinic. This 3-month period is consistent with the practice in previous PSF studies.9 33 56 57 To assess participants’ PSF, a psychiatrist (the principal investigator) who is blinded to the participants’ radiological data will administer the FSS, which is the most frequently used tool to evaluate fatigue in patients who had a stroke.9 33 41 44 56 58–63 The FSS contains nine items (eg, ‘I am easily fatigued’, ‘Exercise brings on my fatigue’ and ‘My motivation is lower when I am fatigued’) that are rated on a 7-point Likert scale, with a higher score indicating greater fatigue. The answers are then averaged, resulting in a global fatigue score. For patients with neurological diseases, the Cronbach’s α of the FSS ranges from 0.90 to 0.94, the intraclass correlation coefficient varies from 0.73 to 0.93 and the correlation coefficients with other fatigue scales vary between 0.62 and 0.84, indicating good internal consistency, reliability and convergent validity, respectively.61 PSF is defined as an FSS Score≥4.0.64 65 PSF severity will be defined by the FSS total score at P1.

A research assistant will administer the Barthel Index (BI),66 MoCA, anxiety subscale of the Hospital Anxiety Depression Scale (HADSA), Geriatric Depression Scale (GDS) and Pittsburgh Sleep Quality Index (PSQI)67 68 to measure the participants’ levels of disability, global cognitive function, anxiety, depressive symptoms and sleep disturbances, respectively. The anxiety subscale of the HADSA contains seven items, each of which is rated on a 4-point Likert scale. Higher scores indicate higher levels of anxiety. The Chinese version of the HADSA, which will be used in the planned study, has been validated.69 The GDS contains 15 items, and the maximum score is 15. It too has been validated in the Chinese population.70 The PSQI contains 19 items, each of which is rated on a scale of 0–3. It measures seven sleep components, and the total score ranges from 0 to 21.

Assessment at follow-ups

For all patients with PSF at P1, two follow-up assessments will be conducted at 9 (P2) and 15 (P3) months post stroke at a research clinic. Similar to the baseline assessment, the FSS, BI, HADSA, GDS and MoCA will be administered at each follow-up assessment. Patients will be divided into remitters and non-remitters. Remission of PSF at P2 or P3 will be defined as a 50% reduction in baseline FSS Score. Additionally, important vascular and clinical events that may affect the PSF outcome, such as recurrent stroke and antidepressant treatment, will be recorded.

MRI data acquisition

Participants will undergo MRI examinations within 2 weeks of the 3-month poststroke assessment using a 3.0 Tesla MRI scanner with a 64-channel receive-only head coil (Prisma, Siemens Healthcare, Germany) using the Human Connectome Project Aging protocol.71 Briefly, diffusion MRI will be performed using the diffusion-weighted spin-echo echo-planar imaging sequence (b-values of 1500 and 3000 s/mm² along 92 diffusion-encoding directions, multiband factor=4, repetition time/echo time=3230/89 ms, field of view=210 mm, 1.5 mm isotropic, 2 acquisitions along positive and negative phase-encoding directions). Resting-state fMRI will be acquired using a gradient-echo echo-planar imaging sequence (multiband factor=8, repetition/echo time=800/37 ms, flip angle=52^o, field of view=208 mm, 2 mm isotropic, number of dynamics=488, 2 acquisitions along positive and negative phase-encoding directions). Participants will be instructed to rest with eyes open during the resting-state fMRI experiment. Structural MRI includes a T1-weighted magnetisation-prepared rapid gradient echo, T2-weighted sampling perfection with application optimised contrast using different flip angle evolution, susceptibility-weighted imaging (SWI) and fluid-attenuated inversion recovery (FLAIR).

Image processing

MRI data, that is, all structural, diffusion and functional MRI images, will be processed using the same preprocessing and postprocessing steps as those described in the Human Connectome Project (HCP)72 using the HCP pipelines,73 FMRIB’s software library V.6.0.4 (FSL) and Connectome Workbench V.1.4.2. For resting-state fMRI data, temporal filtering (0.01<f<0.1 Hz) will be performed.

Old infarcts will be defined as lesions equivalent to the signal characteristics of cerebrospinal fluid on the T1-weighted images and measuring more than 3 mm in diameter, and wedge-shaped cortico-subcortical lesions will be regarded as old infarcts. A semiautomatic quantitative method will be used to measure infarct volume. All raw data will be transferred to an in-house workstation, and the number and laterality of old and lacunar infarcts will be evaluated.

CMBs will be defined as small (2–10 mm) hypointense lesions on the SWI, excluding symmetric basal ganglia calcification and flow void artefacts of the pial blood vessels.74 The number and laterality of microbleeds in each location will be recorded.

WMHs will be measured on FLAIR using the semiautomated segmentation method on the aforementioned in-house workstation. Seeds will be dropped onto the structure of interest and then grown automatically to include all connected pixels until the entire structure is outlined. The volume of the outlined pixels is displayed automatically. The parameters will be calculated for each detected WMH area, including the centre of mass coordinates and Euclidian distance to the ventricular system. If the Euclidian distance is less than 10 mm, a WMH is labelled as periventricular and otherwise as a deep WMH. The stereotactic coordinates of the centre of mass will be calculated using a combination of the T1-to-PD/T2 and T1-to-Talairach space registration matrices. The position of the centre of mass relative to that of the major fissures and sulci in the Talairach atlas will then be determined and used to define the lobar (frontal, temporal, parietal or occipital) localisation of each WMH area. The volumes of periventricular and deep WMHs will be estimated using a similar method.

Image analysis

Brain regions in the CSN: The four brain regions belonging to the CSN, namely, middle frontal gyrus, precentral gyrus, caudate and putamen, will be obtained from the Harvard-Oxford cortical and subcortical atlas.75

FC: Standard seed-based approach will be performed to estimate the FC of the CSN. FC will be computed from the coefficient of Pearson’s correlation between the mean fMRI signal of each of the four regions of the CSN (middle frontal gyrus, precentral gyrus, caudate and putamen) and the fMRI signal of all brain voxels. The FC of voxel or region that overlaps with infarct will be set to zero. This spatial map of correlation will be z-transformed using Fisher’s r-to-z transformation and subsequently averaged between two sessions (positive and negative phase-encoding directions) and between the homologous regions, resulting in a single spatial map for each subject and each region of the CSN. The significant FC in the spatial map will be determined using one-sample t-test on a voxel-by-voxel basis (FDR-corrected p<0.05).

Structural connectivity: The fibre orientation density function will be estimated from diffusion MRI data using FSL’s BEDPOSTX. A dense connectome containing the number of fibre tracks between a pair of greyordinates will be subsequently obtained using probabilistic tractography using FSL’s PROBTRACKX. The dense connectome will then be parcellated using the Harvard-Oxford cortical and subcortical atlas. Diffusion metrics, such as FA, mean diffusivity, diffusional kurtosis, etc, of each of the fibre tracts underlying this parcellated connectome will also be estimated.

Infarct/CMB/WMH analysis: To ensure comparability of brain structure volumes between participants, the MRI data of each participant will be transformed from the original space to a common stereotactic space using multiscale affine registration. Brain regions will be automatically segmented from the head MRI data using a brain extraction tool. Tissue classification in the brain will separate grey matter, white matter and cerebrospinal fluid from the brain MRI data. This process will be performed using the supervised k-nearest neighbour classifier for the entire three-dimensional image. An atlas-based approach will be used to accomplish whole-brain segmentation, in which new input data are automatically adjusted, according to the prior atlas intensity model. This method has demonstrated good robustness. The brain atlas76 will be constructed from manually delineated deep brain structures and intracranial regions (such as the CSN) on a single participant. These brain structure labels will then be transformed into input data non-rigidly (using Demon registration). Infarcts, CMBs and white matter lesions will be segmented using a coarse-to-fine mathematical morphology method developed in-house.

Outcome measures

The primary outcomes of the study include the presence of PSF at baseline and associated CSN FC. The secondary outcomes involve the severity of PSF at baseline and the remission of PSF at follow-ups and the corresponding FC.

Statistical analysis

For the primary analysis, the effect of PSF on the FC, structural connectivity and diffusion metrics of CSN of stroke survivors, voxel-wise two-sample t-tests will be performed with FDR correction for multiple comparisons and significance level set at p<0.05. To avoid the influence of possible confounders, the foregoing analysis will be repeated using an analysis of covariance after controlling for age, sex, NIHSS, BI, HADSA, PSQI and GDS. The same statistical analyses will be performed to compare remitters and non-remitters at P2 and P3 to explore the effects of PSF outcome on CSN. A second-level analysis of the fMRI data using FSS Score as a continuous variable will be performed.

One secondary analysis will be performed. To evaluate the association between PSF symptom severity and CSN among patients with PSF at P1, voxel-wise correlations will be conducted between FC, structural connectivity and diffusion metrics versus FSS total score with FDR correction for multiple comparison and significance level set at p<0.05. To avoid the influence of possible confounders, the BI, HADSA and GDS scores will be included as covariates.

Patient and public involvement

Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.