Robust detection of traumatic axonal injury in individual mild traumatic brain injury patients: Intersubject variation, change over time and bidirectional changes in anisotropy
- First Online:
- Cite this article as:
- Lipton, M.L., Kim, N., Park, Y.K. et al. Brain Imaging and Behavior (2012) 6: 329. doi:10.1007/s11682-012-9175-2
- 985 Views
To identify and characterize otherwise occult inter-individual spatial variation of white matter abnormalities across mild traumatic brain injury (mTBI) patients. After informed consent and in compliance with Health Insurance Portability and Accountability Act (HIPAA), Diffusion tensor imaging (DTI) was performed on a 3.0 T MR scanner in 34 mTBI patients (19 women; 19–64 years old) and 30 healthy control subjects. The patients were imaged within 2 weeks of injury, 3 months after injury, and 6 months after injury. Fractional anisotropy (FA) images were analyzed in each patient. To examine white matter diffusion abnormalities across the entire brain of individual patients, we applied Enhanced Z-score Microstructural Assessment for Pathology (EZ-MAP), a voxelwise analysis optimized for the assessment of individual subjects. Our analysis revealed areas of abnormally low or high FA (voxel-wise P-value < 0.05, cluster-wise P-value < 0.01(corrected for multiple comparisons)). The spatial pattern of white matter FA abnormalities varied among patients. Areas of low FA were consistent with known patterns of traumatic axonal injury. Areas of high FA were most frequently detected in the deep and subcortical white matter of the frontal, parietal, and temporal lobes, and in the anterior portions of the corpus callosum. The number of both abnormally low and high FA voxels changed during follow up. Individual subject assessments reveal unique spatial patterns of white matter abnormalities in each patient, attributable to inter-individual differences in anatomy, vulnerability to injury and mechanism of injury. Implications of high FA remain unclear, but may evidence a compensatory mechanism or plasticity in response to injury, rather than a direct manifestation of brain injury.
KeywordsMild traumatic brain injury (mTBI)MRIDiffusion tensor imaging (DTI)Traumatic axonal injury (TAI)Image processing and analysis
Diffusion tensor imaging
Enhanced Z-score microstructural assessment for pathology
Glasgow Coma Scale
Gaussian Random Field
Health Insurance Portability and Accountability Act
Institutional Review Board
Johns Hopkins University
Montreal Neurological Institute
Mild traumatic brain injury
Receiver operating characteristic
Traumatic axonal injury
Traumatic brain injury
Diagnosis of mild traumatic brain injury (mTBI) is typically based on history and examination. Diagnostic criteria include Glasgow Coma Scale (GCS) of 13–15, loss of consciousness not exceeding 20 min, posttraumatic amnesia not exceeding 24 h and absence of neurological deficits (Esselman and Uomoto 1995) (see Messé, et al. in this issue). Animal studies (Povlishock 1992; Crooks 1991; Pettus et al. 1994; Povlishock 1986; Povlishock et al. 1983; Rubovitch et al. 2011; Greer et al. 2011; Spain et al. 2010) indicate that TBI, including mild injury (Povlishock et al. 1983; Rubovitch et al. 2011; Greer et al. 2011; Spain et al. 2010), results in traumatic axonal injury (TAI), the presumptive pathological substrate of clinical deficits seen in humans (Meythaler et al. 2001; Sharp and Ham 2011; Little et al. 2010) (see Bigler and Maxwell 2012). Despite the strong consensus that clinical manifestations of mTBI are a consequence of TAI, widely used diagnostic tests such as CT and MR imaging, generally have not provided evidence of brain abnormalities (Hammoud and Wasserman 2002).
Diffusion tensor imaging (DTI) reveals evidence of TAI in animal models of TBI (e.g., (Mac Donald et al. 2007a, b; Wang et al. 2009)) and in patients, where brain abnormalities detected by DTI are associated with important clinical outcomes (e.g.,(Kraus et al. 2007; Miles et al. 2008; Niogi et al. 2008a)). Recent studies have used DTI to link specific functional impairment after mTBI to injury at specific brain regions (e.g.,(Niogi et al. 2008b; Geary et al. 2010; Little et al. 2010; Levin et al. 2010; Hartikainen et al. 2010; Lipton et al. 2009)). (See Shenton, et al. 2012) These studies compare groups of patients, implicitly assuming that location of injury will be the same across patients. However, the wide variation in the direction and magnitude of forces applied during head injury makes this assumption highly improbable (Kou et al. 2010; Muller et al. 2009) (see Rosenbaum and Lipton 2012). In addition, DTI must be analyzed at the individual subject level to be useful as a patient-oriented diagnostic tool.
We have developed a whole brain assessment of Fractional Anisotropy (FA) optimized for assessment of individual patients, which we have termed “Enhanced Z-score Microstructural Analysis for Pathology” (EZ-MAP), showing that it discriminates mTBI patients from controls at the time of injury (Kim et al. 2011). Herein we apply this method to the assessment of a cohort of mTBI patients at multiple times after injury to directly assess the utility of the approach and the inter-subject variation in TAI pathology that it reveals.
Materials and methods
After Institutional Review Board (IRB) approval, compliance with the Health Insurance Portability and Accountability Act (HIPAA) and written informed consent, subjects were prospectively enrolled, distinct from clinical care.
Inclusion and exclusion criteria
1. 18–67 years of age
1. Prior head injury
2. Emergency department diagnosis of concussion within 2 weeks
2. Hospitalization due to the injury
3. GCS = 13–15
3. Neurodevelopmental or neurological disorder
4. LOC < 20 min
4. Major psychiatric disorder
5. Posttraumatic amnesia <24 h
5. Illicit drug use within 30 days
6. No focal neurologic deficit
6. Skull fracture or abnormal CT
7. English or Spanish proficiency
1. 18–67 years of age
Same as for patients
Thirty control subjects, with age and gender distribution encompassing that of the patients, were recruited through advertisements. Control subjects underwent the same MR imaging protocol as patients. Similarity of the patient and control groups was confirmed with χ2 (gender) and Student t (age) tests. Controls met the same exclusion criteria applied to patients, including (a) history of prior head injury, (b) history of neurologic or psychiatric disease, and (c) history of illicit drug use.
Imaging was performed using a 3.0 T MRI scanner (Achieva; Philips Medical Systems, Best, the Netherlands) with an eight-channel head coil (Sense Head Coil; Philips Medical Systems). T1-weighted whole-head structural imaging was performed using sagittal three-dimensional magnetization-prepared rapid acquisition gradient echo (MP-RAGE; TR/TE = 9.9/4.6 msec; field of view, 240 mm2; matrix, 240 × 240; and section thickness, 1 mm). T2-weighted whole-head imaging was performed using axial two-dimensional turbo spin-echo (TR/TE = 4000/100 msec; field of view, 240 mm2; matrix, 384 × 512; and section thickness, 4.5 mm) and axial two-dimensional fluid-attenuated inversion recovery turbo spin-echo (TR/TE = 1100/120 msec; inversion time, 2800 msec; field of view, 240 mm2; matrix, 384 × 512; section thickness, 4.5 mm; and number of signals acquired, one) imaging. DTI was performed using single-shot echo-planar imaging (TR/TE = 3800/88 msec; field of view, 240 mm2; matrix, 112 × 89; section thickness, 4.5 mm; independent diffusion sensitizing directions, 32; and b = 1000 s/mm2).
Data quality control
Phantom and actual experimental data are checked for signal-to-noise, geometric distortion and other artifacts as well as head motion during the DTI acquisition. Data of poor quality or with significant gross motion are excluded from analysis.
Neuroradiologic image assessment
Two American Board of Radiology certified neuroradiologists independently reviewed MR images of all subjects (patients and controls) in random sequence during a single session, blind to clinical information and group membership (patient or control).
Calculation of diffusion parameter images
The 33 diffusion-weighted image sets (32 diffusion sensitizing directions and the b = 0 s/mm2 image) were corrected for head motion and eddy current effects using an affine registration algorithm. FA was derived from DTI at each voxel using the FMRIB Diffusion Toolbox (Smith et al. 2007).
Quantitative image analysis was performed as follows:
Non-brain voxels were removed from the MP-RAGE and turbo spin-echo images using FMRIB-FSL software (Smith et al. 2004). Each brain volume was inspected section-by-section, and residual non-brain voxels were removed manually.
Echo-planar imaging distortion correction:
Turbo spin-echo images were acquired with the same section thickness, position and orientation as DTI. Distortion correction was accomplished using a nonlinear deformation algorithm to match each echo-planar image to the corresponding co-planar turbo spin-echo image (Lim et al. 2006).
Intermediate rigid-body registration:
Each subject’s turbo spin-echo images were registered to their three-dimensional MP-RAGE volume using the Automated Registration Toolbox three-dimensional rigid-body approach (Ardekani 1995; Ardekani et al. 2005).
Registration to standard space:
The nonlinear registration module of the Automated Registration Toolbox was used to register each subject’s three-dimensional MP-RAGE volume to a standard T1-weighted template (Montreal Neurological Institute atlas; MNI) (Holmes et al. 1998).
Transformation of DTI to standard space:
Using the Automated Registration Toolbox, distortion correction, intermediate rigid-body registration, and standard space registration were applied to the calculated FA maps in a single resectioning operation. Final cubic voxel size was 1 mm3, masked to exclude non-brain voxels from the analysis.
Our spatial normalization procedure has been shown to be robust across subjects (Ardekani 1995; Ardekani et al. 2005). Nonetheless, the results of each registration are critically assessed by viewing each stage of the registration output, with particular assessment of the alignment of (1) brain surface, (2) deep structures including brainstem, corpus callosum and fornix and (3) grey/white margin in both the deep grey structures and at the cortical margin. These landmarks must align within one-two voxel dimensions for the registration to be accepted, although alignment is typically nearly exact.
White matter segmentation:
The fast automated segmentation tool in the FMRIB-FSL package (Smith et al. 2004) was used to generate a white matter mask for the three-dimensional MP-RAGE template brain images. This mask was eroded by 3 voxels, to eliminate locations most at risk of misregistration, and was used to restrict subsequent statistical analysis of FA to white matter voxels.
The Johns Hopkins University (JHU) white matter atlas (Oishi et al. 2010) was adapted for segmentation of white matter subregions. Using FLIRT from the FMRIB-FSL package (Smith et al. 2004), the T1-weighted JHU white matter atlas was registered to the T1-weighted template. The resulting transformation matrix was applied to the white matter segmentation volume of the JHU white matter atlas to bring it into registration with the MNI template used for DTI analysis.
Prior to subsequent voxelwise analyses, multiple linear regression analysis of the effects of age, gender, and education was performed within 30 control subjects. Regression coefficients were applied to the voxels within each patient’s FA image where effects of covariates on individual control FA voxels were significant at p < 0.05 and more than 100 significant voxels formed a contiguous cluster.
Enhanced Z-score (EZ) analysis:
Details of the analysis procedure are provided in the Appendix. Briefly, the whole-brain voxel-wise Z-score, optimized for assessment of individual patients, was used to identify loci with abnormally low or abnormally high FA in each individual patient compared to the control group. This technique has been optimized and validated previously (Kim et al. 2011). The analysis was performed separately for each patient’s spatially normalized (Montreal Neurological Institute (MNI)) brain volume and includes the following steps: (1) The FA volume for each patient is adjusted for demographic covariates (above). (2) Robust estimation of the control population variance is achieved using a bootstrap resampling procedure. (3) The Enhanced Z-score is computed independently for each voxel in a given patient’s FA volume, using the variance determined from step 1. (4) The Enhanced Z-score map thus generated is thresholded to exclude voxels that do not meet criteria for significant deviation from the control distribution. We accept as significant deviations only clusters of voxels which meet two criteria: (a) |EZ|>1.96 for each voxel and (b) p < 0.01 for the cluster size among voxels meeting the initial EZ-score criterion (a); the cluster size threshold is corrected for multiple comparisons.
Patients (n = 34)
Controls (n = 30)
29.9 ± 6.4
36.6 ± 11.9
38.9 ± 13.2
38.1 ± 10.3
34.9 ± 11.5 (19–64)
37.3 ± 11.0 (20–60)
No. of men**
53 % (16/30)
13.1 ± 2.9 (8–19)
17.0 ± 4.4 (7–26)
Frequently observed regions with abnormally low FA within 2 weeks of mTBI
Number of patients out of 34
Superior corona radiata (R)
Anterior corona radiata (R)
Splenium of corpus callosum (L)
Superior corona radiata (L)
Precentral WM (R)
Posterior Limb of internal capsule (R)
Posterior thalamic radiation (L)
Middle occipital WM (R)
Splenium of corpus callosum (R )
Body of corpus callosum (R)
Posterior limb of internal capsule (L)
Precentral WM (L)
Retrolenticular internal capsule (L)
Frequently observed regions with abnormally high FA within 2 weeks of mTBI
Number of patients out of 34
Anterior corona radiata (R)
Superior corona radiata (R)
Body of corpus callosum (R)
Superior corona radiata (L)
Anterior corona radiata (L)
Genu of corpus callosum (R )
Body of corpus callosum (L)
Superior frontal WM (R )
Posterior corona radiata (L )
Putamen (R )
Posterior corona radiata (R )
Putamen (L )
Change in the number of abnormal voxels between acute post-injury period (<2 weeks) and 3 months
Change over time in the number of abnormal voxels (<2 weeks, 3 months, and 6 months)
We employed EZ-MAP to reveal evidence of TAI in individual mTBI patients at a range of times following injury. We thus add to the body of evidence indicating that brain tissue injury occurs after mild head trauma, even when conventional MR images appear normal (see Bigler and Maxwell 2012). Our study breaks important new ground, however, by demonstrating significant inter-individual differences in TAI. Although several brain regions are consistently affected across most patients, the pattern of abnormalities is unique in each individual. This finding may be explained by the interaction of the unique characteristics of each patient and the particular biomechanical features of each injury. The fact that certain brain areas (e.g., corpus callosum) are commonly abnormal likely reflects a greater susceptibility of these structures to TBI, as described in prior studies of diffuse axonal injury (e.g., McArthur et al. 2004). Even within these susceptible structures, however, the pattern, extent and magnitude of the abnormalities are variable. Variation of mTBI pathology between patients should be expected. Consideration of inter-individual variation may improve diagnosis and prognosis based on DTI and provide a useful individualized proxy endpoint for future clinical trials of TBI treatment. We emphasize that the intersubject variation seen here would be left undetected in a group-wise analysis. See Rosenbaum and Lipton 2012, which discusses the issue of intersubject variation in mTBI.
We chose to study FA because it has been the diffusion parameter most widely explored and has yielded the greatest number of findings in TBI. Several studies have supported the use of FA for identifying white matter abnormalities in mTBI (Miles et al. 2008; Rutgers et al. 2008; Inglese et al. 2005) (see Shenton et al. 2012). FA has also been found to be robust for detection of axonal pathology in experimental studies (Bennett et al. 2012; Mac Donald et al. 2007c) (See Bigler and Maxwell 2012). Nonetheless, the EZ-MAP approach could be applied to maps of axial, radial or mean diffusivity, as well as other quantitative imaging parameters; this is an important area for future investigation.
Areas of abnormally low FA consistent with TAI were detected in almost all patients within 2 weeks of injury (32/34 patients). This finding is consistent with animal and human studies reporting the pathological substrates of diffusion anisotropy and imaging features of mTBI (e.g.,(Mac Donald et al. 2007c)). In white matter, water diffuses more readily parallel to axons because its diffusion in other directions is restricted by subcellular structure including neurofilaments, microtubules, myelin and the axolemma. Intra-axonal microstructural disturbances and degradation of the myelin sheath have been demonstrated using DTI, in the absence of frank axotomy (Song et al. 2003). The shear forces exerted on an axon during even mild head trauma have been reported to cause axonal pathology, with or without ultimate axotomy (Povlishock and Katz 2005) (see Bigler and Maxwell 2012). The study patients sustained mild head injury and as far as it is possible to know, they have no other reasons to have white matter disease. The cohort was carefully screened to exclude confounding variables and no patients had any visible abnormalities on conventional imaging. Our findings thus underscore the fact that mild head trauma causes actual brain injury and that it is detectable at the single subject level.
The abnormally high FA detected in the great majority (32/34 patients) of our patients within 2 weeks of injury is a particularly interesting finding. Biophysically, this phenomenon is unexpected because the loss or disruption of white matter microstructure by mTBI would be expected to manifest as low FA. However, in our study and several others, high FA in the corpus callosum has been detected 72 h (Bazarian et al. 2007), 6 days (45) and 2 weeks (Mayer et al. 2010) after mTBI. These few studies that have found and discussed findings of high FA have assessed patients close to the time of injury and generally put forth the explanation that axonal swelling in the acute post-injury period, due to an intracellular influx of water (i.e., cytotoxic edema), leads to restriction of diffusion within the extracellular space, resulting in increased anisotropic diffusion and therefore higher FA. However, a study of chronic mTBI patients also showed high FA in the internal capsule (Lo et al. 2009). A recent and elegant DTI and histological study of TBI in a rodent model demonstrated increase in anisotropy, which correlated with reactive astrogliosis (Budde et al. 2011). While this association raises additional possibilities for the mechanisms that may lead to increases in anisotropy and warrants further exploration, we note that the finding was limited to grey matter, as seen with both histology and DTI. White matter anisotropy was found to decline in association with markers of axonal injury and was not associated with astrogliosis. It thus remains unclear how gliosis might impact white matter anisotropy. Additionally, the injury model (controlled cortical impact) produced gross tissue disruption and at least a moderate degree of injury, features not present in human mTBI.
The time course of evolving high FA in the present study, though heterogeneous, provides some intriguing clues to pathophysiology. The finding of high FA in all patients at 3 and 6 months post-injury, as well as in chronic symptomatic mTBI patients (see below), is inconsistent with cytotoxic edema. Though present at 2 weeks, the number of voxels with high FA increased from 2 weeks to 3 months in most patients (11 of 16) and then declined from 3 months to 6 months in many (5 of 7). This profile suggests that elevated FA might represent a biophysical manifestation of a response to brain injury, rather than a direct manifestation of injury pathology.
Increased FA may reflect neuroplastic responses to injury, perhaps through up-regulation of active axoplasmic transport. Notably, local increases in anisotropy have been reported in training experiments, where the brain substrate is presumed to be plasticity (Scholz et al. 2009). Compensatory mechanisms may be most active during the acute and sub-acute periods when the opportunity for repair and recovery is maximal. Our observation of decreasing number of low FA voxels is consistent with repair of TAI, while the early increase in or maintenance of number of high FA voxels at 3 months, followed by a decrease in the number of high FA voxels at 6 months post-injury is consistent with the early development of plasticity and other compensatory mechanisms. Further study, including the differential contribution of specific Eigenvalues to abnormally elevated anisotropy may help to clarify the biological role of abnormally high FA.
The finding of high FA in nearly every patient begs the question as to why high FA has not been detected in most prior DTI studies of mTBI. Prior studies employed group-wise analyses, which are inherently insensitive to variability in the spatial location of abnormalities between patients. We have shown much variability in the spatial distribution of high FA voxels between patients, which would likely be missed by a group-wise approach.
Several limitations of this study must be considered. Voxelwise analysis approaches, particularly in individual subjects, may be prone to false positive results. This is because many simultaneous statistical tests are performed. Vulnerability to false positive inference is perhaps the central limitation of such approaches. We therefore take many steps to minimize the chance of such type 1 errors. Two levels of thresholding are employed, such that initially any voxel must meet a threshold (α1) to potentially be identified as an abnormality. Next, contiguous subsets of these voxels are identified as abnormalities only if the size of the cluster of such voxels meets a second threshold (α2). This second threshold is determined based on GRF and is corrected for multiple comparisons. The rationale for the dual threshold approach is that while multiple testing increases the chance for type 1 error at the individual voxel level, these chance errors should be randomly distributed throughout the brain volume. It is not statistically plausible that such random errors will cluster in a contiguous region of the size we require for a final determination of abnormality. It is important to note that the results reported here are based on thresholds (α1, α2) shown to maximize discrimination of patients and controls. At these thresholds, the chance of finding a single suprathreshold cluster is less than 1 %. However, it is important to recognize that sensitivity and specificity vary based on the selected thresholds. Depending on the clinical or research question, a different threshold level might be more appropriate. Specifically, the use of a stricter threshold reduces the chance of a false positive inference (i.e., misclassifying a normal subject as a patient), though at the expense of sensitivity.
A voxelwise analysis provides powerful means for surveying the entire brain for abnormalities and the only means for appreciating the scope of inter-individual differences in the spatial distribution of pathology. An additional advantage of the approach is that it is not subject to the observer bias inherent in the manual delineation of ROIs or tractograms. Nonetheless, the spatial normalization prerequisite to a voxelwise analysis is a potential source of error. This is especially so if distortion is present in the original diffusion-weighted images due to eddy current or magnetic susceptibility-related effects. Images were corrected for the effects of eddy currents and we employed a validated method to correct for distortion prior to image analysis. To ensure that registration of different image types (DTI and MP-RAGE) as well as registration of images from individual subjects would be as accurate as possible, each subject’s eddy current and motion corrected DTI images was registered to their own T2-weighted FSE images, which were subsequently registered to their own high-resolution T1-weighted images and, finally, to a high-resolution T1-weighted template (the MNI atlas brain). We specifically employ a locally implemented template that has been extensively tested to ensure excellent registration of subject data. This approach minimizes the potential for error in inter-modality inter-subject registration. The registration approach employed has been compared to several other methods including AIR, AFNI and SPM (Ardekani 1995; Ardekani et al. 2005) and performs equal to or better than all. Another issue inherent in the spatial normalization process is that the native resolution of the DTI images is coarse relative to the higher resolution structural images. During spatial normalization, DTI voxels are resampled to this higher resolution. It is important to recognize that the native voxel thus interpolated to multiple smaller voxels does not support inference regarding this finer spatial scale. Ultimately, what we can learn from the data remains limited to effects detectable on the scale of the native DTI resolution.
It is well known that FA is reduced at locations where white matter fibers cross within a voxel [e.g., (Poupon et al. 2000)]. Such regions of low FA due to fiber crossing could possibly be detected as pathologically abnormal FA. However, since crossing fibers tend to occur in the same location across subjects, a systematic difference in crossing fiber anatomy between a patient and the control group is not likely and therefore not likely to be detected as abnormal, although where a single patient is compared to a group of normals, the risk for this problem may be somewhat greater than in a group vs. group comparison.
As is a common problem in longitudinal studies, particularly in patients with variable actual and perceived impairment, we encountered significant attrition during follow-up. This limits our sample size at the three and six month follow-up assessments. Additionally, while we do not have specific reason to suspect bias, is possible that attrition during follow-up could lead to selection bias, were patients either more or less severely injured or impaired selectively more or less likely to comply with follow-up.
Controls and patients reported different educational experience, which was significant. Although it would be ideal to control for this variable during subject recruitment, as we did with age and gender, this difference has been accounted for using standard approaches. In addition, the effect of education on FA was tested and found to be quite modest (Kim et al. 2011).
In conclusion, EZ-MAP detects multifocal abnormally low and high FA, consistent with TAI and, we propose, plasticity or other compensatory processes, respectively. This effective individual level assessment of mTBI yields patient-specific information. Such individual measurements can be used to assess the relationship of each patient’s injury to their functional outcome, which is an essential step toward determining the utility of DTI as a prognostic biomarker, which might serve to guide future personalized medicine approaches to the treatment of mTBI.