Longitudinal Brain Magnetic Resonance Imaging CO2 Stress Testing in Individual Adolescent Sports-Related Concussion Patients: A Pilot Study

Background Advanced neuroimaging studies in concussion have been limited to detecting group differences between concussion patients and healthy controls. In this small pilot study, we used brain magnetic resonance imaging (MRI) CO2 stress testing to longitudinally assess cerebrovascular responsiveness (CVR) in individual sports-related concussion (SRC) patients. Methods Six SRC patients (three males and three females; mean age = 15.7, range = 15–17 years) underwent longitudinal brain MRI CO2 stress testing using blood oxygen level-dependent (BOLD) MRI and model-based prospective end-tidal CO2 targeting under isoxic conditions. First-level and second-level comparisons were undertaken using statistical parametric mapping (SPM) to score the scans and compare them to an atlas of 24 healthy control subjects. Results All tests were well tolerated and without any serious adverse events. Anatomical MRI was normal in all study participants. The CO2 stimulus was consistent between the SRC patients and control subjects and within SRC patients across the longitudinal study. Individual SRC patients demonstrated both quantitative and qualitative patient-specific alterations in CVR (p < 0.005) that correlated strongly with clinical findings, and that persisted beyond clinical recovery. Conclusion Standardized brain MRI CO2 stress testing is capable of providing a longitudinal assessment of CVR in individual SRC patients. Consequently, larger prospective studies are needed to examine the utility of brain MRI CO2 stress testing as a clinical tool to help guide the evaluation, classification, and longitudinal management of SRC patients.

Background: Advanced neuroimaging studies in concussion have been limited to detecting group differences between concussion patients and healthy controls. In this small pilot study, we used brain magnetic resonance imaging (MRI) CO2 stress testing to longitudinally assess cerebrovascular responsiveness (CVR) in individual sports-related concussion (SRC) patients.
Methods: Six SRC patients (three males and three females; mean age = 15.7, range = 15-17 years) underwent longitudinal brain MRI CO2 stress testing using blood oxygen level-dependent (BOLD) MRI and model-based prospective end-tidal CO2 targeting under isoxic conditions. First-level and second-level comparisons were undertaken using statistical parametric mapping (SPM) to score the scans and compare them to an atlas of 24 healthy control subjects.
results: All tests were well tolerated and without any serious adverse events. Anatomical MRI was normal in all study participants. The CO2 stimulus was consistent between the SRC patients and control subjects and within SRC patients across the longitudinal study. Individual SRC patients demonstrated both quantitative and qualitative patient-specific alterations in CVR (p < 0.005) that correlated strongly with clinical findings, and that persisted beyond clinical recovery.
conclusion: Standardized brain MRI CO2 stress testing is capable of providing a longitudinal assessment of CVR in individual SRC patients. Consequently, larger prospective studies are needed to examine the utility of brain MRI CO2 stress testing as a clinical tool to help guide the evaluation, classification, and longitudinal management of SRC patients.
Keywords: sports-related concussion, adolescent, post-concussion syndrome, magnetic resonance imaging, blood oxygen level-dependent imaging, cerebrovascular reactivity, longitudinal Longitudinal CVR Assessment Following SRC Frontiers in Neurology | www.frontiersin.org July 2016 | Volume 7 | Article 107 inTrODUcTiOn Concussion is a form of traumatic brain injury (TBI) that represents an important public health concern, especially in children and adolescents participating in sports. Although the vast majority will recover within 1-4 weeks of injury, an important proportion will develop prolonged symptoms or post-concussion syndrome (PCS) (1)(2)(3)(4)(5)(6). This heterogeneous group of patients can be classified clinically into those with exercise intolerance, vestibulo-ocular dysfunction, cervical spine soft tissue injury, migraine headaches, and post-injury psychiatric disorders (7)(8)(9)(10)(11)(12). In recent years, several advanced neuroimaging techniques have demonstrated significant differences in white matter anisotropy, brain activation patterns, and resting cerebral blood flow (CBF) between groups of concussion patients and healthy control subjects (13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23). Although these studies have advanced our understanding of the effects of concussion on brain structure and function, examination of group differences in neuroimaging biomarkers may not be an accurate reflection of the pathophysiological mechanisms underlying the symptoms and injury states in individual sports-related concussion (SRC) patients.
Consequently, there remains a persistent need for neuroimaging assessment tools that can inform the evaluation and longitudinal management of individual concussion patients. One of the metrics defining the health of the system responsible for regulating CBF during states of rest, activity, and disease is cerebrovascular reactivity/responsiveness (CVR), which is defined as the change in CBF in response to a vasodilatory stimulus (24,25). Although previous studies have used techniques, such as transcranial Doppler ultrasonography to investigate changes in CVR following concussion (26)(27)(28), magnetic resonance imaging (MRI)-based assessment tools have several advantages, including superior spatial resolution and the ability to evaluate the entire brain (29)(30)(31). However, to provide accurate and longitudinal MRI-based assessment of CVR, a quantifiable, reliable, and reproducible vasoactive stimulus must be paired with an MRI sequence that can measure CBF or its surrogate (32,33). Using model-based prospective targeting (MPET) of CO2 applied to blood oxygen level-dependent (BOLD) MRI, our group has introduced a brain MRI CO2 stress test that is capable of providing whole brain CVR mapping in individual concussion patients (34,35). In a recent study, we demonstrated group and individual differences in CVR among adolescent PCS patients compared with healthy control subjects (35).
In this pilot study, we report results of longitudinal qualitative and quantitative brain MRI CO2 stress testing in individual adolescent SRC patients, comparing them to an atlas of healthy adolescent control subjects tested in the same way.

MaTerials anD MeThODs research Design
This study was approved by the Biomedical Research Ethics Board at the University of Manitoba. We conducted a prospective case-control study of adolescent SRC patients and healthy control subjects. All adolescent SRC patients were recruited from the Pan Am Concussion Program in Winnipeg, MB, Canada.
Patient inclusion criteria for this study included (1) diagnosis of SRC according to the definition set forth by the International Consensus on Concussion in Sport (36); (2) age 13-21 years. Healthy control subjects were recruited through word of mouth, including patient siblings and relatives. Control subject inclusion criteria included age 25 or younger. Control subject exclusion criteria were as follows: (1) the presence of a symptomatic concussion; (2) history of prior TBI or neurological condition resulting in structural brain abnormality on previous neuroimaging; (3) contra-indication to MRI; (4) presence of a neurological condition requiring prescription medication.

clinical assessment
All adolescent SRC patients underwent a clinical evaluation by a single neurosurgeon. All healthy control subjects underwent clinical interview to collect demographic data, past medical history, and past concussion history. In general, patients were deemed completely recovered when they were asymptomatic, tolerating full-time school, and neurological examination results were normal. In-season athletes were also required to successfully complete the International Consensus on Concussion in Sport return-to-play guidelines (36) to be deemed completely recovered. In some cases, hybrid neuropsychological testing and graded aerobic treadmill testing were used to confirm clinical recovery as indicated by the treating neurosurgeon. Patients were diagnosed with physiological post-concussion disorder if they had symptoms for 1 month or longer and demonstrated a symptom-limited threshold on graded aerobic treadmill testing (8,11). Patients were diagnosed with vestibulo-ocular post-concussion disorder if they demonstrated subjective and objective evidence of vestibulo-ocular dysfunction symptoms for 1 month or longer and had no clinical evidence of exercise intolerance (9). Patients were diagnosed with a post-injury psychiatric disorder if they developed a novel psychiatric disorder following a SRC (10).

neuroimaging assessment
Following informed consent and clinical assessment, all subjects underwent neuroimaging assessment. SRC patients underwent initial neuroimaging assessment during the symptomatic phase of their injury and follow-up neuroimaging after documented clinical recovery. Follow-up neuroimaging assessment for SRC patients was scheduled based on patient and parent convenience rather than a pre-determined time point. MPET CO2 targeting was achieved by precise delivery of CO2 using a sequential gas delivery breathing circuit connected to a computerized gas-blender (RespirAct, Thornhill Research Inc., Toronto, ON, Canada). This device allows precise manipulation of PETCO2 levels under isoxic (target PETO2 = 115 mmHg) conditions. Hemodynamic monitoring during the study period included continuous heart rate and pulse oximetry and non-invasive blood pressure measurement (BP) at 3-min intervals.

Structural Neuroimaging
The structural neuroimaging component of each study was reviewed by a board-certified neuro-radiologist.

Preprocessing of MRI Sequences
Standard preprocessing of MRI EPI output was carried out for the BOLD sequences using SPM8 software and included batch processing by an SPM toolbox and custom written in-house MatLab scripts. The BOLD data were interpolated to the MPRAGE voxel dimensions. The fMRI model specification was a 2-pass process. We first modeled the data using the finite impulse response (FIR) package in SPM. The hypercapnic MPET model, as described earlier, using a triple stimulus block design for CO2 square wave response was assessed with a zero offset for this stimulus. The whole brain response to the CO2 stimulus was examined. An event-related response was calculated. The time to maximal response to hypercapnia was noted. This delay in response to the change in end-tidal CO2 was corrected based on a constructed series of time delay block stimulus files generated with delays from 0 to 30 s. This new fMRI model based on the time delay for brain activation by CO2 was now rerun as above.
We used the motion correction file generated with realignment as regressors in the model. First-level analysis results were based on this time-corrected analysis and the contrast images generated here were used in the second-level analysis. Masks were generated to correct for known inhomogeneous output of BOLD EPI signals. We made individualized masks based on GRE B0-field inhomogeneities greater than and less than 500 arbitrary units (AU). These binary images masked out voxels with inhomogeneities at the base of the brain, those adjacent to the petrous bone and those contiguous with the frontal sinuses. A dilated CSF mask with dilation × 2 (based on the Wake Forest University Pick Atlas) was generated to mask out the ventricular and periventricular inverse BOLD response in part related to choroid plexus response and interface changes at the ventricular edges (37). This combined binary image -labeled [(gm + wm) − (B0_inhomogeneity + dilated_ventricles)] was used in the second-level analysis as an inclusive mask to ensue that the abnormal voxels for an individual study were intraparenchymal. A representative mask is shown in the Supplementary Material. First-level analyses were undertaken with the author blinded to the subject's group (healthy control versus SRC patient); however, second-level analyses were not blinded since they were based on the results of the first-level analysis.

Cerebrovascular BOLD Responsiveness (First-Level Analysis)
First-level analyses were undertaken for each study participant. BOLD increases in response to the hypercapnic breathing stimulus and BOLD decreases in response to the stimulus were assessed at the p = 0.001 uncorrected level. The cluster size threshold was 10 voxels. The response to the applied CO2 stimulus and its inverse were expressed as voxel counts/whole-brain voxel count ratios.

Cerebrovascular BOLD Responsiveness (Second-Level Analysis)
A second-level analysis for the BOLD studies was completed based on comparison to an atlas of adolescent control subjects (n = 24), thereby allowing individual longitudinal analysis of the six SRC patients. Each SRC patient study underwent voxelby-voxel comparisons for BOLD signals that were less than or greater than the mean control group response. This second-level analysis was conducted over a series of p-values (p = 0.001, 0.005, 0.01, and 0.05; we report our results at the p = 0.005 level; see below). For each individual second-level comparison, the images were masked using the combined mask outlined above for that individual versus the atlas output, and the voxel counts reported are corrected for the applied individual mask.

Participants
In this study, longitudinal brain MRI CO2 stress testing was carried out in six SRC patients (three males and three females; mean age = 15.7, range = 15-17 years) whose test results were compared with those of a normal control atlas of 24 subjects (15 males and 9 females; mean age = 18.5, range = 13-25 years). Four SRC patients and 17 control subjects participated in a previous cross-sectional study (35). Three control subjects were excluded due to excessive motion during neuroimaging assessment. Among control subjects that made up the normal control atlas, their past medical history was significant for remote concussion in 7/24 (29.2%) subjects. One SRC patient was on medication for a post-injury anxiety and sleep disorder during both the initial and follow-up neuroimaging. Additional clinical and neuroimaging characteristics of the SRC patients are summarized in Table 1.

study Tolerability
All subjects successfully completed the full imaging study. One SRC patient reported transient and self-limiting headache and fatigue during initial neuroimaging assessment.

structural neuroimaging
Structural neuroimaging studies were normal in all subjects. No evidence of traumatic abnormalities was detected among SRC patients.

BOlD cerebrovascular responsiveness (First-level analysis)
There was no significant difference between the control subject atlas and SRC patient BOLD responses to CO2 (p = 0.001; 10 voxel cluster size). The mean BOLD response to the CO2 stimulus was 83 ± 12% in the SRC patients and 85 ± 6% in the control subjects (p = 0.668 between groups). The mean BOLD inverse response was 0.5 ± 0.4% in the SRC patients and 0.5 ± 0.4% in controls (p = 0.774 between groups). The number of voxels masked out by the inclusive mask [(gm + wm) − (B0_inhomogeneity + dilated_ventricles)] was 20,980 ± 3900 in the SRC group at imaging period 1 and 21,290 ± 4400 at imaging period 2, indicating no difference in number of voxels masked out between time periods (p = 0.919). The proportion of voxels masked out by the B0 field inhomogeneities represented 5.4 ± 1.7% of total voxel counts during imaging period 1 and 4.5 ± 1.7% in imaging period 2. Dilated ventricle masking represented 5.7 ± 3.2 and 6.9 ± 2.1%, respectively.

BOlD cerebrovascular responsiveness (second-level analysis)
Second-level analysis comparing individual SRC patients to the healthy control atlas at their initial neuroimaging tests, during the symptomatic phase of injury, demonstrated patient-specific alterations in CVR in all patients with a predominant pattern of increased CVR. Follow-up neuroimaging, during the clinically recovered stage of injury, demonstrated relatively stable CVR in two patients, and improved but persistently abnormal CVR in four patients. Examples of longitudinal CVR assessment using brain MRI CO2 stress testing is shown in Figures 1 and 2. Images indicate voxel counts at the p = 0.005 level. The full complement of patient study images is available in Supplementary Material.  Figure 1 during the symptomatic (a) and clinically recovered stages of injury (B) compared with the stand-alone control atlas. Results are reported at the p = 0.005 level for cut-off of significant voxels (t-statistic value = 2.82 in this case). Hot color voxels (orangish hues) indicate where BOLD CVR response is significantly greater than the mean value for the collated control group images and the cool color voxels (blue hues) indicate where the BOLD CVR response is significantly less when compared with the mean value for the control group. The scale indicates the color scheme as expressed in t-values. This approach permits quantitation of the longitudinal differences evident to the eye as seen in Figure 1 to be displayed. The results of this pilot study, therefore, provide confirmatory evidence that SRC patients display both qualitative and quantitative patient-specific alterations in CVR that can be detected by brain MRI CO2 stress testing. Our findings also suggest that alterations in CVR may account for some of the clinical findings in patients with acute concussion and physiological post-concussion disorder, but that CVR changes may play a less prominent role in those with isolated vestibulo-ocular dysfunction and postinjury psychiatric disorders. Our results are in agreement with previous work that demonstrated alterations in mean resting CBF in groups of pediatric and collegiate SRC patients that persisted beyond clinical and neurocognitive recovery (18,23). Whether alterations in resting CBF and CVR that persist beyond clinical recovery place these athletes at an elevated risk of future injury, or lead to long-term effects of cumulative brain injury, requires further study. It is possible that the vessels marked by abnormal CVR may, like other tissues, be more vulnerable to re-injury, or worsened physiological dysfunction, with repeated head trauma. The persistence of CVR abnormalities detected months after injury in many of these patients indicates that much longer evaluations are required to document the natural history of cerebrovascular dysfunction following concussion. Our preliminary experience with this population suggests that concussion may be characterized by a predominant pattern of increased CVR during the acute or sub-acute phase and a predominantly reduced pattern of CVR in the chronic phase. If CVR is an adequate biomarker for cerebrovascular dysfunction following concussion, additional study will be required to further refine the criteria for abnormal CVR and the adequacy of neurovascular recovery as well as take into account potential artifact related to data analysis and fluctuations in CVR over time among normal subjects.
Brain MRI CO2 stress testing meets several important requirements for a neuroimaging assessment tool that can provide longitudinal CVR assessments in concussion patients. First, as demonstrated here and in previous studies (34,35) this test is safe and well tolerated among acute SRC and PCS patients and does not require exposure to radiation, provocative breathing maneuvers, or intravenous agents. Second, this technique delivers a precise, quantifiable, and reproducible CO2 stimulus under isoxic conditions, thereby allowing a consistent vasoactive stimulus to be applied during serial assessments without the confounding effects of alterations in PETO2 and individual variability in respiratory physiology (32,38). Third, this technique generates qualitative and quantitative biomarkers that can be compared with other clinical outcomes in individual SRC patients (33). Although there were subtle 1-2 mmHg differences in the CO2 stimulus applied across subjects and time-points, this level of end-tidal gas control is similar to that achieved with mechanical ventilation in other CVR studies (39,40), and represents a level of precision and reproducibility that is unachievable with other vasoactive stimuli including breath-holding, acetazolamide, and inhaled CO2 such as carbogen (32,33).
This pilot study has several important limitations. First, the number of SRC patients who underwent longitudinal CVR assessment is small. Patients were recruited from a tertiary pediatric concussion program that may have selected for more severely injured SRC patients. Second, objective measures of clinical concussion recovery, including graded aerobic treadmill testing and formal neuropsychological testing, were not used to confirm physiological and neurocognitive recovery in all patients, and patients were not imaged at uniform and pre-selected time periods. Some patients had returned to contact and collision sports prior to follow-up neuroimaging assessment in which cases the effects of unreported concussions or sub-concussion injuries may have affected follow-up imaging results. Third, masking of the BOLD EPI for the second-level analysis removed from consideration some of the prefrontal and frontal cortex and periventricular white matter in each case that may have resulted in under-detection of abnormal voxels. Lastly, although previous work has demonstrated excellent within-day reliability of CVR assessment in pediatric subjects using this technique (41), longitudinal CVR assessments were not performed among the healthy control subjects. This limitation should be considered in future studies.
In conclusion, this study provides confirmatory empirical evidence that SRC is associated with patient-specific alterations in CVR. Larger prospective studies are needed to confirm these preliminary findings and examine the utility of brain MRI CO2 stress testing as a clinical tool to help guide the evaluation, classification, prognostication, and longitudinal management of SRC patients.
aUThOr cOnTriBUTiOns WM -experimentation, data collation, analysis, interpretation, and writing. MJE -experimentation, data collation, analysis, interpretation, and writing. LR -experimentation, data collation, analysis, interpretation, and writing. MM -experimentation and data collation. PP -experimentation and data collation. BD -analysis and statistical consultation. ME -analysis and interpretation. DM -analysis, interpretation, and writing. JD -analysis, interpretation, and writing. JF -analysis, interpretation, and writing.

FUnDing
All phases of this study were supported by grants from the Health Sciences Center Foundation, the Manitoba Health Research Council and the Academic Oversight Committee, Winnipeg, MB, Canada.