Detecting axonal injury in individual patients after traumatic brain injury

Abstract

Poor outcomes after traumatic brain injury (TBI) are common yet remain difficult to predict. Diffuse axonal injury is important for outcomes, but its assessment remains limited in the clinical setting. Currently, axonal injury is diagnosed based on clinical presentation, visible damage to the white matter or via surrogate markers of axonal injury such as microbleeds. These do not accurately quantify axonal injury leading to misdiagnosis in a proportion of patients. Diffusion tensor imaging provides a quantitative measure of axonal injury in vivo, with fractional anisotropy often used as a proxy for white matter damage. Diffusion imaging has been widely used in TBI but is not routinely applied clinically. This is in part because robust analysis methods to diagnose axonal injury at the individual level have not yet been developed. Here, we present a pipeline for diffusion imaging analysis designed to accurately assess the presence of axonal injury in large white matter tracts in individuals. Average fractional anisotropy is calculated from tracts selected on the basis of high test-retest reliability, good anatomical coverage and their association to cognitive and clinical impairments after TBI. We test our pipeline for common methodological issues such as the impact of varying control sample sizes, focal lesions and age-related changes to demonstrate high specificity, sensitivity and test-retest reliability. We assess 92 patients with moderate-severe TBI in the chronic phase (≥6 months post-injury), 25 patients in the subacute phase (10 days to 6 weeks post-injury) with 6-month follow-up and a large control cohort (n = 103). Evidence of axonal injury is identified in 52% of chronic and 28% of subacute patients. Those classified with axonal injury had significantly poorer cognitive and functional outcomes than those without, a difference not seen for focal lesions or microbleeds. Almost a third of patients with unremarkable standard MRIs had evidence of axonal injury, whilst 40% of patients with visible microbleeds had no diffusion evidence of axonal injury. More diffusion abnormality was seen with greater time since injury, across individuals at various chronic injury times and within individuals between subacute and 6-month scans. We provide evidence that this pipeline can be used to diagnose axonal injury in individual patients at subacute and chronic time points, and that diffusion MRI provides a sensitive and complementary measure when compared to susceptibility weighted imaging, which measures diffuse vascular injury. Guidelines for the implementation of this pipeline in a clinical setting are discussed.

Keywords: diagnostic pipeline; diffuse axonal injury; diffusion tensor imaging; traumatic brain injury.

Figure 1

DTI diagnostic pipeline processing steps. 1: Skeletonized FA maps are derived for the patient of interest and the comparative control group using TBSS. 1.1: If a patient has evidence of contusion or missing brain, these areas are manually delineated using in house software (ImSeg) to produce a lesion mask. The lesion mask is then registered to standard MNI152 1 mm space and subtracted from tract masks so that these regions are excluded from influencing subsequent FA measures. 2: Skeletonized FA maps from controls and individual patients are tract masks to extract mean FA value for each region of interest (ROI) for each individual. 3: Control FA distributions are assessed for assumptions of normality using histograms, QQ plots and Shapiro-Wilks tests. 3.1: If assumptions of normality are violated, rank-inverse normalization is applied. 4: Control group mean FA and SD for each region of interest is then calculated for use in Z-scoring. 5: Z-scores for each patient are calculated. 6: Z-scores for a patient are then converted to P-values (95% CI, two tailed) and corrected for multiple comparisons using FDR. Whole brain FA skeleton is calculated as a stand-alone test. Individual patient data are then plotted against control group and significantly lower FA in patients are highlighted in red for ease of interpretation.

Figure 2

Data-driven tract selection. (A) Voxel size of all 46 candidate tracts ordered from largest to smallest. Median and first interquartile range demonstrated on red dashed lines. (B) Mean FA in each of the remaining candidate tracts across a longitudinal healthy control population (n = 20). (C) Reliability of tracts calculated using ICC using longitudinal data from healthy control subset. See Supplementary Table 3 for tract abbreviations and associated acronyms.

Figure 3

Tracts selected for use in diagnostic pipeline. Left: Rendering of tracts selected for use in the DTI pipeline. The corpus callosum demonstrates the three subdivisions as: red = splenium, blue = body, and orange = genu. L = left; R = right. Right: Estimation plots representing individual mean FA values for TBI patients (PAT) and controls (CON). Black lines represent group level means. A measure of effect size and its 95% CIs calculated through non-parametric bootstrap resampling are plotted on the vertical bar on a separate but aligned axes (Ho et al., 2012).

Figure 4

Diagnostic results using the DTI diagnostic pipeline for each individual TBI patient. (A) Example of individual patient diagnostic results across all tracts and whole brain skeleton. Individual points denote the individual patient FA. Box plots represent the distribution of mean FA for a given tract across all healthy controls (n = 103). A red circle indicates a significantly abnormal FA value for the individual patient compared to controls (P < 0.05, 95% CI, FDR corrected). A black circle indicates a patient's mean FA value fell within normal ranges compared to healthy controls (P < 0.05, 95% confidence interval, FDR corrected). (B) Percentage of patients identified as having an abnormality in tract regions of interest and whole brain skeleton as quantified by diagnostic pipeline P-value (two-tailed, 95% CI, FDR corrected). All categories = total percentage of abnormalities across all damage categories. (C) Number of tract abnormalities per patient identified as abnormal using diagnostic pipeline compared to 103 healthy controls. Colours represent the damage category of a patient, e.g. yellow indicates where a TBI patient has tract abnormalities but no evidence of damage on routine clinical MRI sequences. CCB/G/S = body/genu/splenium of the corpus callosum; CR_L/R = left/right corona radiata; CST_L/R = left/right corticospinal tract; ILF_L/R = left/right inferior longitudinal fasciculus; MCP = middle cerebellar peduncle; WSKL = whole brain white matter skeleton.

Figure 5

Individual case studies. Routine MRI scans including T₁-MPRAGE, SWI and FLAIR of four moderate-severe TBI patients assessed using the DTI pipeline are observed on the top row of the case study boxes. Areas of visible contusion or microbleeds are highlighted by red circles and confirmed in radiological reports by a consultant neuroradiologist. The second row of the case studies box demonstrates the patients individual DTI diagnostic results. Box plots represent FA of the control cohort used in the pipeline (n = 103). Single points on the plot represents the individual patient assessed with the pipeline. Black dots represent where a patient does not have a significantly abnormal FA within a given region of interest. Red points indicate where patients have a significantly abnormal FA compared to the healthy control cohort, P < 0.05, 95% CI, FDR corrected. Bottom row of the case studies illustrate individual patients’ cognitive performance (yellow dot) compared to the control cohort (n = 35) highlighted as blue box plots. Lower scores in the memory and reasoning domains indicate poorer performance whilst higher scores in the executive and information processing domains indicate poorer performance due to slower reaction times. B/G/SCC = body/genu/splenium of the corpus callosum; CR_L/R = left/right corona radiata; CST_L/R = left/right corticospinal tract; ILF_L/R = left/right inferior longitudinal fasciculus; MCP = middle cerebellar peduncle; WSKL = whole brain white matter skeleton.

Figure 6

Reliability and accuracy measures using smaller control group sample sizes to identify DTI abnormalities in individual TBI patients. (A) ICC values derived from the FDR corrected P-values of individual patient’s tract FA when compared to randomly subsampled control groups of a given n over 100 iterations. Sample sizes ranged from 10 to 50 and were randomly selected from the total control cohort (n = 103). (B) Calculation of type 1 and type 2 errors and related measures of sensitivity, specificity, positive and negative predictive value for diagnostic rates of individual TBI patients when compared to a control group of a given n (e.g. 10–50 in increments of 10). Values are calculated by comparing results of smaller control group to the results derived for each patient when compared to all 103 healthy controls. CC = corpus callosum; CR = corona radiata; CST = corticospinal tract; ILF = inferior longitudinal fasciculus; MCP = middle cerebellar peduncle.

Figure 7

Reliability of diagnostic pipeline across varying control sample sizes. (A) Comparison of composite scores for the four cognitive domains selected between healthy controls and patients identified as having either normal or abnormal DTI using the DTI pipeline. Diagnosis of normal versus abnormal DTI was determined using the whole control cohort (n = 103), not accounting for age. (B) Comparison of functional outcome measures between patients identified as having normal versus abnormal DTI using the diagnostic pipeline. ***P < 0.001, *P < 0.05.

Figure 8

Longitudinal diagnostic results of subacute TBI cohort (n = 25). (A) Number of abnormal tracts identified in longitudinal subacute TBI cohort using diagnostic pipeline for subacute and chronic time points. (B) Tract fractional anisotropy measures from subacute (10 days to 6 weeks) and chronic (∼6 months) time points. B/G/SCC = body/genu/splenium of the corpus callosum; CR_L/R = left/right corona radiata; CST_L/R = left/right corticospinal tract; ILF_L/R = left/right inferior longitudinal fasciculus; MCP = middle cerebellar peduncle; WSKL = whole brain white matter skeleton.