Blood-Brain Barrier Disruption in Mild Traumatic Brain Injury Patients with Post-Concussion Syndrome: Evaluation with Region-Based Quantification of Dynamic Contrast-Enhanced MR Imaging Parameters Using Automatic Whole-Brain Segmentation.

OBJECTIVE: This study aimed to investigate the blood-brain barrier (BBB) disruption in mild traumatic brain injury (mTBI) patients with post-concussion syndrome (PCS) using dynamic contrast-enhanced (DCE) magnetic resonance (MR) imaging and automatic whole brain segmentation.
MATERIALS AND METHODS: Forty-two consecutive mTBI patients with PCS who had undergone post-traumatic MR imaging, including DCE MR imaging, between October 2016 and April 2018, and 29 controls with DCE MR imaging were included in this retrospective study. After performing three-dimensional T1-based brain segmentation with FreeSurfer software (Laboratory for Computational Neuroimaging), the mean Ktrans and vp from DCE MR imaging (derived using the Patlak model and extended Tofts and Kermode model) were analyzed in the bilateral cerebral/cerebellar cortex, bilateral cerebral/cerebellar white matter (WM), and brainstem. Ktrans values of the mTBI patients and controls were calculated using both models to identify the model that better reflected the increased permeability owing to mTBI (tendency toward higher Ktrans values in mTBI patients than in controls). The Mann-Whitney U test and Spearman rank correlation test were performed to compare the mean Ktrans and vp between the two groups and correlate Ktrans and vp with neuropsychological tests for mTBI patients.
RESULTS: Increased permeability owing to mTBI was observed in the Patlak model but not in the extended Tofts and Kermode model. In the Patlak model, the mean Ktrans in the bilateral cerebral cortex was significantly higher in mTBI patients than in controls (p = 0.042). The mean vp values in the bilateral cerebellar WM and brainstem were significantly lower in mTBI patients than in controls (p = 0.009 and p = 0.011, respectively). The mean Ktrans of the bilateral cerebral cortex was significantly higher in patients with atypical performance in the auditory continuous performance test (commission errors) than in average or good performers (p = 0.041).
CONCLUSION: BBB disruption, as reflected by the increased Ktrans and decreased vp values from the Patlak model, was observed throughout the bilateral cerebral cortex, bilateral cerebellar WM, and brainstem in mTBI patients with PCS.

INTRODUCTION

Post-concussion syndrome (PCS) following mild traumatic brain injury (mTBI) is a major health and socioeconomic burden worldwide, especially in developed countries (1). Specifically, among 1.7 million people per year who visited hospitals in the United States for traumatic brain injury (TBI), an estimated 75% or more had mild injuries, according to the 2007 Centers for Disease Control and Prevention report (23). Despite the high incidence and clinical importance of mTBI, the role of imaging has been neglected in the evaluation of mTBI than in moderate TBI or TBI of greater severity (34). One of the reasons for this is the lack of a standardized and optimal imaging modality to evaluate mTBI accurately.

Over the past few years, there have been several attempts to evaluate mTBI noninvasively using various MR sequences, most of which were based on susceptibility-weighted imaging (SWI) (5678) and diffusion tensor imaging (DTI) (91011) for detecting microhemorrhages and fiber tract injuries, respectively. In line with these attempts, dynamic contrast-enhanced (DCE) MR imaging, an advanced MR imaging technique that can quantify the extent of blood-brain barrier (BBB) disruption in various central nervous system diseases through pharmacokinetic modeling (1213141516), has been highlighted as a potentially useful tool to evaluate mTBI in several preclinical studies (171819).

Moreover, the potential clinical utility of DCE MR imaging has also been investigated in mTBI patients with PCS, as conventional MR imaging performed in case of persistent PCS symptoms often fails to reveal any abnormality, causing frustration to both clinicians and patients in routine practice. A previous study on football players with histories of concussion revealed BBB dysfunction in various cortical regions (20). Furthermore, a more recent study on mTBI patients with PCS demonstrated that DCE MR imaging parameters, based on a two-compartment pharmacokinetic model proposed by Tofts and Kermode (122122), significantly differed between mTBI patients and controls in certain manually defined regions-of-interest (ROIs) (23), implying that BBB disruption may have an important role in the pathophysiology of PCS in mTBI patients. However, given that manual ROI placement can be subjective, and subtle BBB disruption may be better reflected by the Patlak model (122122), the potential clinical utility of DCE MR imaging in mTBI patients with PCS, as demonstrated by the previous study (23), remains elusive.

Therefore, this study aimed to further investigate BBB disruption in mTBI patients with PCS using DCE MR imaging and automatic whole-brain segmentation.

MATERIALS AND METHODS

This retrospective study was approved by the Institutional Review Board of our hospital, and the need for informed consent was waived.

Patient Selection

Based on our radiology report database between October 2016 and April 2018, 45 consecutive patients were identified under the following inclusion criteria: 1) patients with a clinical diagnosis of mTBI based on loss of consciousness not exceeding more than 0–30 minutes, posttraumatic amnesia or alteration of consciousness for less than 24 hours (24), 2) patients with a clinical diagnosis of PCS as defined by the International Classification of Diseases, 10th Revision (25), and 3) patients with contrast-enhanced MR imaging, including DCE MR imaging. Clinical diagnoses of mTBI and PCS were arrived at after history taking and neuropsychological examinations (Rivermead Post-concussion symptoms Questionnaire, RPQ [n = 30] and computerized neurocognitive function tests, CNTs [n = 26]) by the rehabilitation physician, blinded to the results of the DCE MR imaging (262728) (Supplementary Table 1 for specific PCS symptoms in mTBI patients. Details for RPQ and CNTs are provided in Supplementary Materials). Exclusion criteria were as follows: 1) patients with severe structural abnormalities, including overt traumatic hemorrhage (n = 2), or 2) failure of automatic brain segmentation (n = 1) (Fig. 1).

Fig. 1

Flowchart for selecting the study population.

DCE = dynamic contrast-enhanced, ICD-10 = International Classification of Diseases, 10th Revision, mTBI = mild traumatic brain injury, PCS = post-concussion syndrome

For control subjects, 35 patients, who had undergone brain MR imaging, including DCE MR imaging, between November 2016 and March 2018, were identified under the following inclusion criteria: 1) patients with various neurological symptoms (headache [n = 25], mild memory impairment [n = 2], visual disturbance [n = 1], and dizziness [n = 1]), 2) patients with contrast-enhanced MR imaging, including DCE MR imaging, 3) patients with normal findings on conventional MR imaging, except for a few T2 hyperintensities not exceeding the age threshold (24), and 4) patients with no prior trauma history. Six patients were excluded owing to failure of automatic brain segmentation (Fig. 1).

Finally, 71 study subjects (42 patients and 29 controls) were included in this study. We reviewed the medical records of the mTBI patients and controls for the presence of various comorbidities, such as diabetes, hypertension, dyslipidemia, chronic kidney disease, cerebral amyloid angiopathy, smoking history, and coronary artery disease.

Automatic Brain Segmentation and DCE MR Imaging Analysis

All MR images were acquired with a 3T scanner (Discovery 750, GE Healthcare) using a 32-channel head coil. The MR sequences included pre- and post-contrast three-dimensional (3D) fast spoiled gradient-echo (FSPGR), T2 fluid-attenuated inversion recovery (3D CUBE), diffusion-weighted imaging, SWI, and DCE MR imaging. For the DCE sequence, 3D fat-suppressed FSPGR imaging was obtained after intravenous administration of gadobutrol (at a dose of 0.1 mmol/kg of body weight), followed by a 30 mL saline bolus, at a rate of 4 mL/s using a power injector (Spectris, MedRad). Forty images were obtained at intervals equal to the repetition time for each section, resulting in a total acquisition time of 5 minutes and 8 seconds. Specific DCE MR imaging parameters were as follows: repetition time, 3 msec; echo time, 1.1 msec; flip angle, 10°; field of view, 240 × 240; matrix, 128 × 128; section thickness, 3 mm; and number of signals acquired, 0.7 (imaging parameters for other MR sequences are provided in Supplementary Table 2).

Based on the pre-contrast 3D T1-weighted images, automatic cortical reconstruction and volumetric whole-brain segmentation were performed using open-source software (FreeSurfer, version 6.0, Laboratory for Computational Neuroimaging). Using a dedicated commercial software package (NordicICE, version 4.1.2, NordicNeuroLab), Ktrans and vp maps were derived based on two pharmacokinetic models (the Patlak model and extended Tofts and Kermode model) by two reviewers (with 3 and 9 years of experience in neuroradiology, respectively), blinded to the results of neuropsychological tests, according to previously described methods (1323). Specifically, DCE MR imaging parameters were obtained using the arterial input function (AIF), which was semiautomatically derived from the main intracranial arteries, including the middle cerebral arteries at the level of the circle of Willis. This process was performed with the consensus of two radiologists (with 3 and 9 years of experience in neuroradiology, respectively). The final AIF curve was obtained using the cluster analysis technique. Subsequently, brain masks at five different regions (i.e., bilateral cerebral cortex and white matter [WM], bilateral cerebellar cortex and WM, and brainstem) were extracted (Fig. 2) and automatically co-registered with the Ktrans and vp maps to estimate mean Ktrans and vp values of these regions. Of note, Ktrans values of the mTBI patients and controls were calculated using both models to identify the model that better reflected the increased permeability owing to mTBI (tendency toward higher Ktrans values in mTBI patients than in controls).

Fig. 2

Automatic brain segmentation and ROI mask extraction.

Based on precontrast three-dimensional T1-weighted images (A, C), automatic cortical reconstruction and volumetric whole-brain segmentation were performed, and brain masks at bilateral cerebral cortex (yellow, B), bilateral cerebral WM (red, B), bilateral cerebellar cortex (yellow, D), bilateral cerebellar WM (red, D), and brainstem (brown, D) were extracted afterward. ROI = region-of-interest, WM = white matter

Statistical Analysis

The normality of the parameters was assessed using the Kolmogorov-Smirnov test. The Wilcoxon signed-rank test was used to compare the mean Ktrans values between the Patlak model and extended Tofts and Kermode model. The Student's t test (or Mann-Whitney U test) was subsequently performed to compare DCE MR imaging parameters and noncategorical clinical variables between mTBI patients and controls. Subgroup analysis was performed according to the time interval between injury and MR imaging (three months or less vs. longer than three months) to compare the DCE MR imaging parameters between the two groups. Categorical clinical variables were compared between the two groups using Fisher's exact test. Pearson correlation analysis or Spearman's rank correlation test was used, as appropriate, to identify the correlations between the DCE MR imaging parameters and clinical parameters, including the time interval between the injury and MR imaging and neuropsychological test scores. Specifically, we assessed the correlation between DCE MR imaging parameters and neuropsychological test scores only for the subgroup of patients with a time interval of ≤ 2 weeks between the neuropsychological tests and MR scanning. The DCE MR imaging parameters of patients with average or good performance in CNTs were further compared to those with moderately or markedly atypical performance using the Mann-Whitney U-test. Receiver operating characteristic (ROC) curves were obtained to evaluate the diagnostic performance of the DCE MR imaging parameters and their optimal threshold values to discriminate between patients with average or good performance in CNTs and those with moderately or markedly atypical performance. Intraclass correlation coefficients (ICC) were calculated to evaluate interobserver agreements for DCE MR imaging parameters. All statistical analyses were performed using MedCalc, version 11.1.1.0 (MedCalc) and SPSS Statistics for Windows, Version 23.0 (IBM Corp.). A p value of less than 0.05 was considered statistically significant.

RESULTS

Demographic and Conventional MR Findings of mTBI Patients and Controls

Age did not significantly differ between mTBI patients {median, 47 years (interquartile range [IQR], 36–53 years)} and controls (median, 52 years [IQR, 43–57 years]) (p = 0.064). No statistically significant difference in sex was found between the two groups (p = 0.177). In addition, there was no significant difference in the incidence of comorbidities between these two groups (11.9% [5/42] vs. 6.9% [2/29]; p = 0.692) (Table 1).

Table 1

Clinical Characteristics of mTBI Patients and Controls

Characteristics	mTBI (n = 42)	Controls (n = 29)	P
Age (y)*	47 (36–53)	52 (43–57)	0.064
Sex			0.599
Male	13 (31.0)	7 (24.1)
Female	29 (69.0)	22 (75.9)
Time interval between injury and MR imaging (months)*	2.0 (1.0–5.0)	NA	NA
Time interval between MR imaging and neuropsychological tests (weeks)*
RPQ (n = 30)	1.7 (0.8–3.3)	NA	NA
CNT (n = 26)	0.7 (0–1.7)	NA	NA
Comorbidity	5 (11.9)	2 (6.9)	0.692
Diabetes mellitus	0 (0)	1 (3.4)	0.408
Hypertension	2 (4.8)	0 (0)	0.510
Dyslipidemia	3 (7.1)	1 (3.4)	0.640
Chronic kidney disease	0 (0)	0 (0)	NA
Cerebral amyloid angiopathy	0 (0)	0 (0)	NA
Smoking	0 (0)	0 (0)	NA
Coronary artery disease	0 (0)	0 (0)	NA

Unless otherwise indicated, data represent the number of patients. Data in the parentheses are percentages. *Data are reported as medians (interquartile range). CNT = computerized neurocognitive function test, mTBI = mild traumatic brain injury, RPQ = Rivermead Post-concussion symptoms Questionnaire, NA = not available

Among the 71 study subjects, including mTBI patients, none showed discernable trauma-related findings, such as intracranial hemorrhage or fracture on conventional MR sequences. However, most of the patients (69.0% [29/42] of the mTBI group, 58.6% [17/29] of the control group) had a few foci with T2 high signal intensity scattered in the cerebral WM not exceeding the age threshold (p = 0.451).

DCE MR Imaging Analysis

Comparison of Mean Ktrans according to Pharmacokinetic Models

Table 2 shows the mean Ktrans values based on two different pharmacokinetic models in the mTBI and control groups. The mean Ktrans values based on the Patlak model were significantly lower than those based on the extended Tofts and Kermode model at all five locations (p < 0.001). Increased permeability owing to mTBI was observed in the Patlak model, but not in the extended Tofts and Kermode model (Fig. 3).

Table 2

Mean Ktrans according to Pharmacokinetic Models

Brain Regions	mTBI (n = 42)			Controls (n = 29)
	Patlak	Extended TK	P	Patlak	Extended TK	P
Bilateral cerebral cortex (x 10-1 min-1)	0.0104 (0.0084–0.0135)	0.1389 (0.1015–0.1815)	< 0.001	0.0084 (0.0072–0.0119)	0.1452 (0.1243–0.1697)	< 0.001
Bilateral cerebral WM (x 10-1 min-1)	0.0036 (0.0028–0.0046)	0.0468 (0.0397–0.0649)	< 0.001	0.0031 (0.0027–0.0041)	0.0470 (0.0414–0.0559)	< 0.001
Bilateral cerebellar cortex (x 10-1 min-1)	0.0081 (0.0062–0.0112)	0.1131 (0.0875–0.1558)	< 0.001	0.0074 (0.0056–0.0089)	0.1380 (0.1036–0.1656)	< 0.001
Bilateral cerebellar WM (x 10-1 min-1)	0.0041 (0.0033–0.0049)	0.0568 (0.0425–0.0799)	< 0.001	0.0038 (0.0027–0.0043)	0.0659 (0.0555–0.0824)	< 0.001
Brainstem (x 10-1 min-1)	0.0052 (0.0044–0.0062)	0.0660 (0.0515–0.0961)	< 0.001	0.0856 (0.0657–0.1021)	0.0856 (0.0657–0.1021)	< 0.001

Data represent medians (interquartile range) unless otherwise noted. Extended TK = extended Tofts and Kermode model, WM = white matter

Fig. 3

Bar graphs show mean Ktrans values according to pharmacokinetic models.

A tendency toward higher Ktrans values in mTBI patients than in controls was observed in the Patlak model but not in the Extended TK. Each error bar depicts the 95% confidence interval for the median. Extended TK = extended Tofts and Kermode model

Comparison of DCE MR Imaging Parameters (from the Patlak Model) between mTBI Patients and Controls

All patients showed symmetrical perfusion on the visual assessment of both Ktrans and vp maps. However, the mean Ktrans value of the bilateral cerebral cortex was significantly higher in mTBI patients (median, 0.0104 × 10−1 min−1 [IQR, 0.0084–0.0135 × 10−1 min−1]) than in controls (median, 0.0084 × 10−1 min−1 [IQR, 0.0072–0.0119 × 10−1 min−1]) (p = 0.042) (Table 2, Figs. 4, 5). Meanwhile, the mean vp values of the bilateral cerebellar WM and brainstem were significantly lower in mTBI patients than in controls (bilateral cerebellar WM: median, 0.85 [IQR, 0.73–0.99] vs. median, 1.00 [IQR, 0.84–1.16], p = 0.009; brainstem: median, 1.15 [IQR, 1.01–1.43] vs. median, 1.40 [IQR, 1.13–1.64], p = 0.011) (Table 3, Figs. 4, 6).

Fig. 4

Notched box-and-whisker plots for DCE MR imaging parameters at bilateral cerebral cortex, bilateral cerebellar WM, and brainstem.

Notched lines of boxes represent median values, while boundaries of the box show the first and third interquartile values. Lower and upper ends of the plots denote the minimal and maximal values and outliers beyond this range are displayed individually.

A. Mean Ktrans value in bilateral cerebral cortex was significantly higher in mTBI patients (median, 0.0104 × 10−1 min−1) than in controls (median, 0.0084 × 10−1 min−1). B, C. mTBI patients had significantly lower mean vp values at bilateral cerebellar WM and brainstem (0.85 and 1.15, respectively) than controls (1.00 and 1.40, respectively).

Fig. 5

Representative Ktrans maps before and after co-registration with ROI masks for bilateral cerebral cortex in a 43-year-old mTBI female patient (A) and a 42-year-old female control (B).

(A, B; upper rows) Ktrans maps before co-registration demonstrate apparently higher Ktrans values in the mTBI patient than in the control throughout bilateral cerebral cortex. (A, B; lower rows) Mean Ktrans values measured from the co-registered Ktrans maps were 0.0261 × 10−1 min−1 for the mTBI patient and 0.0045 × 10−1 min−1 for the control.

Table 3

Comparison of DCE MR Imaging Parameters (from the Patlak Model) between mTBI Patients and Controls

Brain Regions	mTBI (n = 42)	Controls (n = 29)	P
Bilateral cerebral cortex
Mean Ktrans (x 10-1 min-1)	0.0104 (0.0084–0.0135)	0.0084 (0.0072–0.0119)	0.042
Mean vp	1.93 (1.71–2.25)	1.98 (1.80–2.35)	0.433
Bilateral cerebral WM
Mean Ktrans (x 10-1 min-1)	0.0036 (0.0028–0.0046)	0.0031 (0.0027–0.0041)	0.277
Mean vp	0.76 (0.64–0.85)	0.75 (0.66–0.90)	0.575
Bilateral cerebellar cortex
Mean Ktrans (x 10-1 min-1)	0.0081 (0.0062–0.0112)	0.0074 (0.0056–0.0089)	0.175
Mean vp	1.50 (1.34–1.81)	1.70 (1.48–2.13)	0.107
Bilateral cerebellar WM
Mean Ktrans (x 10-1 min-1)	0.0041 (0.0033–0.0049)	0.0038 (0.0027–0.0043)	0.198
Mean vp	0.85 (0.73–0.99)	1.00 (0.84–1.16)	0.009
Brainstem
Mean Ktrans (x 10-1 min-1)	0.0052 (0.0044–0.0062)	0.0052 (0.0043–0.0066)	0.991
Mean vp	1.15 (1.01–1.43)	1.40 (1.13–1.64)	0.011

Data represent medians (interquartile range) unless otherwise noted. All the DCE parameters were derived using the Patlak model. DCE = dynamic contrast-enhanced

Fig. 6

Representative vp maps before and after co-registration with ROI masks for bilateral cerebellar WM and brainstem in a 41-year-old mTBI female patient (A) and a 43-year-old female control (B).

(A, B; upper rows) vp maps before co-registration depict apparently lower vp values in the mTBI patient than in the control throughout the cerebellar WM and brainstem. (A, B; lower rows) Mean vp values in bilateral cerebellar WM measured from the co-registered vp maps were 0.48 for the mTBI patient and 1.16 for the control, while mean vp values in the brainstem were 0.85 for the mTBI patient and 1.44 for the control.

Correlations between DCE MR Imaging Parameters (from the Patlak Model) and Time Interval between Injury and MR Imaging

DCE MR imaging parameters (i.e., the mean Ktrans of the bilateral cerebral cortex, vp of the bilateral cerebellar WM, and vp of the brainstem) were not significantly correlated with the time interval between injury and MR imaging (ρ = 0.028, p = 0.858; ρ = −0.090, p = 0.571; ρ = −0.081, p = 0.612; respectively).

In the subgroup analysis according to the time interval between injury and MR imaging (three months or less vs. longer than three months), the mean Ktrans values of the bilateral cerebral cortex of both subgroups tended to be higher than those of controls, although the difference was not statistically significant (for three months or less, p = 0.093; for longer than three months, p = 0.077) (Supplementary Table 3).

The mean vp values of the bilateral cerebellar WM of both subgroups were significantly lower than those of controls (for three months or less, p = 0.047; for longer than three months, p = 0.012). The mean vp values of the brainstem of both subgroups tended to be lower than those of controls, although the statistical significance was only achieved for the subgroup with the interval longer than three months (for three months or less, p = 0.086; for longer than three months, p = 0.008) (Supplementary Table 3).

Correlation between DCE MR Imaging Parameters (from the Patlak Model) and Neuropsychological Tests

Detailed scores for all RPQ (n = 30) and CNTs (n = 26) are summarized in Table 4. Among those, 18 patients and 19 patients had imaging-neuropsychological test time intervals of two weeks or less for RPQ and CNTs, respectively. In this subgroup of patients, no significant correlation was found between DCE MR imaging parameters and neuropsychological test scores (all p > 0.05) (Fig. 7). However, the mean Ktrans value of the bilateral cerebral cortex was significantly higher in patients with atypical performance in the auditory continuous performance test (commission errors) (median, 0.0122 × 10−1 min−1 [IQR, 0.0096–0.0165 × 10−1 min−1]) than in those with average or good performance (median, 0.0089 × 10−1 min−1 [IQR, 0.0080–0.0111 × 10−1 min−1]) (p = 0.041). In the ROC analysis, the mean Ktrans value of the bilateral cerebral cortex had a sensitivity of 70.0% and a specificity of 88.9% for differentiating the patients with atypical performance and those with average or good performance at a cut-off value of 0.0093 × 10−1 min−1.

Table 4

Scores of RPQ and Neurocognitive Function Tests

Tests	Scores
RPQ* (n = 30)	34.0 (27.0–45.0)
RPQ-3	6.5 (5.0–9.0)
RPQ-13	27.0 (21.0–37.0)
CNT (n = 26)
Auditory CPT (correct responses)†	40.0 (27.0–49.0)
Auditory CPT (commission errors)†	45.0 (27.0–62.0)
VLT (immediate recall)†	52.5 (45.0–59.0)
VLT (delayed recall)†	50.0 (42.0–70.0)
VLT (delayed recognition)†	46.5 (36.0–57.0)
Digit span test (forward)†	43.0 (35.0–46.0)
Digit span test (backward)†	50.0 (38.0–53.0)
Card sorting test (perseverative response)†	51.5 (46.0–64.0)

Data are reported as the median (interquartile range). *The total RPQ score for 16 PCS symptoms ranges from 0 to 64 with a higher score representing a greater degree of symptoms. The test is composed of RPQ-3 (3 items for scoring headaches, nausea and/or vomiting, and dizziness, representing early concussion symptoms) and RPQ-13 (13 items for scoring cognitive, mood, sleep, and other physical symptoms, indicative of later symptoms of PCS) (26), †T scores. CPT = continuous performance test, PCS = post-concussion syndrome, VLT = verbal learning test

Fig. 7

A heatmap for the Spearman correlation analysis between DCE MR imaging parameters and neuropsychological tests.

No significant correlation was found between DCE MR imaging parameters and neuropsychological test scores (all p > 0.05). CPT = continuous performance test, RPQ = Rivermead Post-concussion symptoms Questionnaire, VLT = verbal learning test

Interobserver Agreement for Quantitative DCE MR Imaging Parameters (from the Patlak Model)

The ICC value for interobserver agreement was almost perfect (Ktrans value, 0.916 [95% confidence interval, CI]: 0.899–0.930; vp value, 0.988 [95% CI]: 0.985–0.991) between the two observers.

DISCUSSION

BBB disruption has received increased attention as one of the major pathophysiologies underlying TBI, based on the results of previous studies that assessed BBB disruption in animal models using Evans blue staining (21) or immunohistochemistry for serum proteins (29). In parallel with the growing evidence from histopathological data, DCE MR imaging has been highlighted as a promising imaging counterpart for the noninvasive evaluation of BBB injury in TBI animal models (2130) and humans (2023). To our knowledge, this study is one of the few clinical studies applying DCE MR imaging on mTBI patients. Unlike previous studies (2031), we applied automatic segmentation and two different pharmacokinetic models (the Patlak model and extended Tofts and Kermode model) for a more rigorous analysis of the subtle alteration of BBB permeability in mTBI. We found that the Ktrans value of the cerebral cortex was significantly higher in mTBI patients compared to the controls. Meanwhile, the vp values of the cerebellar WM and brainstem were lower in the patients than in the controls. Furthermore, the Ktrans value of the cerebral cortex was significantly higher for poor performers on the auditory continuous performance test (commission errors).

In DCE MR imaging, contrast media composed of low-molecular gadolinium reach the brain tissue and distribute between the blood plasma and extravascular extracellular space (EES) through the BBB, which acts as a filter. Under the circumstances of BBB injury following trauma, gadolinium distribution across the BBB differs from that in normal brain tissue. This difference in gadolinium distribution owing to the change in BBB permeability can be calculated based on a DCE MR pharmacokinetic model and represented by quantitative variables, such as Ktrans (i.e., volume transfer constant between the plasma and EES), vp (i.e., fractional plasma volume), and ve (i.e., fractional interstitial volume). There are several pharmacokinetic models in use, with the conventional Tofts model, extended Tofts model, and Patlak model being the most popular (122122). Both the extended Tofts model and the Patlak model can be used for highly perfused tissue. However, unlike the extended Tofts model, which considers bidirectional transport, including back-diffusion from the EES into plasma, the Patlak model ignores this back-diffusion and therefore provides estimates for only two parameters (Ktrans and vp). Because of the additional complexity of the extended Tofts model in considering back-diffusion, which may cause over-fitting in low-permeability settings, the Patlak model is reported to be more accurate for measuring normal and slightly increased permeability with negligible back-diffusion (122132). Although we found measured Ktrans values to be significantly higher on the extended Tofts and Kermode model than in the Patlak model, our results demonstrated that the Patlak model might be more suitable for reflecting the difference between mTBI patients with low permeability and controls, in keeping with the previous findings.

Based on DCE imaging parameters obtained using the Patlak model, we demonstrated that the BBB permeability of the cerebral cortex was significantly different between mTBI patients and controls, in keeping with a previous study that reported BBB lesions in various cortical regions in football players with histories of concussion (20). Several studies have demonstrated cortical involvement in mTBI patients. Specifically, in brain contusions, which are one of the structural abnormalities found in mTBI, collision between soft brain parenchyma and the surrounding rigid bone usually affects the subpial cortical surface first and subsequently extends through the cortex with disruption of tissue and vessels or BBB (3334). A reduction in cortical thickness was also noted in ice hockey team members with PCS (35).

In addition to the cortical BBB abnormality, vp values were found to be significantly lower in the cerebellar WM and brainstem. Although these regions are known to be more frequently affected in severe injury than in mild injury, diffuse neuronal and axonal injury has also been reported in mTBI patients, as evidenced by a global decrease of WM fractional anisotropy (including the brainstem and cerebellums) on DTI (3637) as well as a reduction in whole-brain N-acetylaspartate (38).

Our study results are similar to those of a previous study, which reported higher Ktrans and ve values at various regions (including T2 hyperintense WM lesions, normal-appearing WM, and predilection sites for diffuse axonal injury) in mTBI patients with PCS than in controls (23). However, we applied automatic whole-brain segmentation to select ROIs rather than manually defining the areas as in the previous study. Therefore, we believe that our results avoid the inherent subjectivity of manual ROI placement and provide more objective evidence for BBB disruption in mTBI patients. Moreover, it is noteworthy that a significant difference was observed in the Ktrans value of the cerebral cortex between poor and good performers in the auditory continuous performance test (commission errors). Although the results are preliminary given the small sample size, the difference may indicate that DCE MR imaging parameters could be imaging biomarkers not only for the BBB injury itself but also for the severity of impairment in attention following mTBI.

Unlike the previous study, which reported high BBB permeability in the cerebral WM (20), we did not find a significant difference in permeability in the bilateral cerebral WM between mTBI patients with PCS and controls. The lack of statistically significant difference may be partially attributed to the use of mean values for the entire cerebral WM, which would lead to averaging the data from not only the site of traumatic impact but also the unaffected WM. We inevitably used the mean values because the coup sites were not obvious from either medical histories or MR images in our patients. Thus, further investigation in future studies, including patients with known coup sites, is warranted.

This study has several limitations apart from those of a retrospective study, including selection bias of the patient pool. First, some of the patients did not undergo neuropsychological tests, and thus, these cases were excluded from the subgroup analysis. In addition, we could not strictly control the time interval between MR imaging and neuropsychological tests, which further limited the study population for the subgroup analysis. These factors led to the small sample size, which could have partially contributed to the lack of significant association between DCE MR imaging parameters and neuropsychological test scores other than auditory continuous performance test (commission errors). However, it was inevitable, given the clinical feature of mTBI with PCS, which is characterized by fluctuating symptoms with variable onsets that do not require prompt medical management, limited resource availability (especially MR scanners), as well as the retrospective nature of the study. Therefore, a further prospective study with larger sample size is needed to validate the correlation. Second, the control group without TBI was not explicitly age- and sex-matched. Nevertheless, there were no significant differences in age and sex between mTBI patients and controls in the statistical analysis, with the median age being slightly younger in the patients. Considering that the aging process can also increase BBB permeability, higher permeability in mTBI patients despite the younger median age supports the significant influence of trauma on BBB permeability. The detected difference in permeability could have been even higher if age-matched controls had been included. Moreover, given that BBB permeability can be increased by numerous neurological diseases, such as primary headache (e.g., migraine) (13), the possibility remains that the inclusion of age-matched individuals without any neurological symptoms, as controls, could have resulted in even lower Ktrans and higher vp for the control group, which may further amplify the detected differences between the patients and controls. Third, caution is warranted when assessing DCE MR imaging parameters (especially vp) because several factors, such as plasma perfusion, can also affect the absolute values of the parameters resulting in values exceeding the theoretical range (i.e., vp higher than 1 in certain regions). Nonetheless, the observed tendency in permeability difference between the mTBI and control groups would still hold true because we applied the same model for these two groups. Fourth, we used mean values for the entire cerebral cortex rather than those for cortical regions at the site of traumatic impact because the coup sites were not obvious from either medical histories or MR images in our patients. Therefore, the difference in BBB permeability between the mTBI with PCS group and controls could have been underestimated.

In conclusion, DCE MR imaging parameters derived from the Patlak model can depict subtle changes in BBB permeability in mTBI patients with PCS and therefore may have a complementary role in mTBI patients with otherwise normal conventional MR imaging findings. In particular, increased BBB permeability, as reflected by higher Ktrans and lower vp, values, was observed throughout the cerebral cortex, cerebellar WM, and brainstem in mTBI patients with PCS.