Robust detection of traumatic axonal injury in individual mild traumatic brain injury patients: Intersubject variation, change over time and bidirectional changes in anisotropy

Abstract

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.

Appendix: Data analysis procedures

Adjustment for demographic covariate effects

We chose control subjects with an even distribution of age, gender and educational attainment that fully brackets the range of the patients; no patient age or educational attainment exceeds all controls at either extreme. FA images used in subsequent analyses were adjusted by regression coefficients (age, gender, years of education) estimated from control subjects at each voxel. Regression coefficients thus determined were applied to FA images of the patients, but only at locations where effects on individual voxels were significant at p < 0.05 and where more than 100 significant voxels formed a contiguous cluster.

Enhanced Z-score (EZ)

We computed the Z-score defined by ${Z} = {{{({X} - mean)}} \left/ {{SD}} \right.}$ at each voxel (i) within a patient’s FA volume with reference values (mean and Standard Deviation (SD)) computed from the control group. It therefore follows that Z-score of a patient may vary with the composition of the control group, with potential for unreliable inferences when the reference group is small. We employed a bootstrap procedure to overcome this potential for sample-to-sample variation of Z-scores, and calculated EZ-score by $E{Z} = {{{{Z}}} \left/ {{\widehat{\sigma}}} \right.}$, where $\widehat{\sigma}$ is the bootstrap SD estimate of Z-scores at a voxel. We applied two levels of thresholding to identify significantly abnormal voxels. First, each voxel must meet a threshold, |EZ| > 1.96. Second, the subset of these voxels that forms contiguous clusters meeting a size threshold (1 %) based on the Gaussian Random Field (GRF) theory (Friston et al. 1994); the cluster size threshold is corrected for multiple comparisons. These thresholds provide maximal discrimination of patients and controls based on maximal area under the ROC curve (results presented previously (Kim et al. 2011)) determined from a range of thresholds ( voxel z-score: 2.5758 and 1.96; cluster size: 0.01 (corrected), 0.05 (corrected), 0.01 (uncorrected) and 0.05 (uncorrected)). It is important to recognize that some voxels meeting the criteria for abnormality are found in controls when these optimal thresholds are applied. We therefore tested the difference in the numbers of abnormal voxels between patients (n = 34) and unique normal control subjects (n = 21) (i.e., not the same individuals used to compute the reference mean and SD for use in the E-Z calculation). The mean number of abnormal voxels found in patients was significantly greater than that of controls (p = 0.014, 2-tails) (Kim et al. 2011).

One Sample t-Test:

An Enhanced Z-score (EZ_i) of i^th mTBI patient at a voxel (voxel location index is omitted) is written as $E{Z_i} = \frac{{{X_i} - \overline Y }}{{{S_Y}{{\widehat{\sigma }}^B}}}$, where X_i is FA of i^th patient; ($\overline Y, {S_Y}$) is mean and SD of control group, respectively; ${\widehat{\sigma }^B}$is bootstrap SD estimate of Z-scores. T-score for the 1-sample t-Test with null hypothesis that is the mean Enhanced Z-scores of patients would be equal to zero is written as

$$ T = \frac{{\sqrt {{{n_x}}} E\overline Z }}{{SD(EZ)}} = \frac{{\sqrt {{{n_x}}} \left( {\overline X - \overline Y } \right)}}{{SD(EZ){S_Y}{{\widehat{\sigma }}^B}}} = \frac{{\sqrt {{{n_x}}} \left( {\overline X - \overline Y } \right)}}{{{S_X}}} \sim t\left( {{n_x} - 1} \right). $$

(1)

since $SD(E{Z_i}) = \frac{{{S_X}}}{{{S_Y}{{\widehat{\sigma }}^B}}}$ (SD of Enhanced Z-scores of patients), where S_x is SD of FAs from patients, and n_x is the number of patient.