Symptom-Dependent Changes in MEG-Derived Neuroelectric Brain Activity in Traumatic Brain Injury Patients with Chronic Symptoms

Abstract

Neuroelectric measures derived from human magnetoencephalographic (MEG) recordings hold promise as aides to diagnosis and treatment monitoring and targeting for chronic sequelae of traumatic brain injury (TBI). This study tests novel MEG-derived regional brain measures of tonic neuroelectric activation for long-term test-retest reliability and sensitivity to symptoms. Resting state MEG recordings were obtained from a normative cohort, Cambridge Centre for Ageing and Neuroscience (CamCAN), baseline: n = 619; mean 16-month follow-up: n = 253) and a chronic symptomatic TBI cohort, Targeted Evaluation, Action and Monitoring of Traumatic Brain Injury (TEAM-TBI), baseline: n = 64; mean 6-month follow-up: n = 39). For the CamCAN cohort, MEG-derived neuroelectric measures showed good long-term test-retest reliability for most of the 103 automatically identified stereotypic regions. The TEAM-TBI cohort was screened for depression, somatization, and anxiety with the Brief Symptom Inventory and for insomnia with the Insomnia Severity Index. Linear classifiers constructed from the 103 regional measures from each TEAM-TBI cohort member distinguished those with and without each symptom, with p < 0.01 for each-i.e., the tonic regional neuroelectric measures of activation are sensitive to the presence/absence of these symptoms. The novel regional MEG-derived neuroelectric measures obtained and tested in this study demonstrate the necessary and sufficient properties to be clinically useful-i.e., good test-retest reliability, sensitivity to symptoms in each individual, and obtainable using automatic processing without human judgement or intervention.

Keywords: CamCAN; TEAM-TBI; anxiety; depression; insomnia; normative atlas; post-concussion syndrome; somatization; test-retest reliability.

Figure 1

The regions whose neuroelectric activity values contributed to the symptom-specific classifiers reported in Table 1 and Table 2 are shown. Each row shows the regions for the indicated symptom. From top to bottom, they are insomnia, insomnia (2nd step), somatization, depression, and anxiety. The MR imaging (MRI) slices range from left to right, inferior to superior in one-centimeter increments. The left side of each slice is the right brain. Activity in blue/red regions was higher/lower in those who screened positive. The cyan landmarks are the boundaries between gray and white matter in the precentral, cingulate, insula, and fusiform regions.

Figure 2

The classifier scores for each baseline CamCAN subject (blue, n = 589) and each TEAM-TBI subject (red, n = 63) are shown in the left. The score on the cortical/subcortical classifier is plotted on the x-axis; the score on the deep white matter classifier is plotted on the y-axis. These are the classifiers whose accuracies are shown in Table 5. In the right panel, all baseline TEAM-TBI scores are plotted, n = 63. Those who returned for follow-up (n = 40) are plotted as arrows; the baseline is plotted at the base of the arrow; the follow-up is plotted at the arrow-head. 63 age and sex matched CamCAN subjects who returned for follow-up are plotted in blue. The standard deviation bars represent 2.0 standard deviations for the classifier scores.

Figure 3

The regions whose neuroelectric activity values contributed to the cohort-specific classifiers reported in Table 5 and Table 6 are shown. The MRI slices range left to right, inferior to superior, in one-centimeter increments. The left side of each slice is the right brain. Activity in blue/red regions was higher/lower in the TEAM-TBI cohort. The cyan landmarks are the boundaries between gray and white matter in the precentral, cingulate, insula, and fusiform regions.

Figure 4

Regional activity (z-score) for 17 subcortical regions is shown for each CamCAN (upper panel) and TEAM-TBI (lower panel) subject. Since the means and standard deviations for the CamCAN subjects were used to compute the z-scores, the mean for each of the CamCAN regions is zero and the z-scores for each region are distributed approximately normally about zero. The mean and Scheme 3. 0 or z > 3.0—i.e., p < 0.0014. Individuals whose measures fall into those portions of the graph that are highlighted in gray are far from the norm. Note that assessing sub-mean normality is limited by the floor.

Figure 5

Regional activity (z-score) is shown for each TEAM-TBI subject for 34 right cortical regions. As in Figure 2, each horizontal bar represents the floor below which no measure can go. The right middle temporal and lateral occipital cortices, indicated with the wide black bars, contributed significantly to the classifier. Note that the means for some regions are greater than zero, some less. See the Figure 4 legend for details.

Figure 6

Short-term (same sitting) vs. long-term (mean 16-months) test-retest reliability is plotted for each of the 103 cortical, subcortical, and deep white matter regions. Regional measures (z-scores) of tonic neuroelectric activity were compared across all CamCAN subjects. Comparisons were between baseline resting and task recordings (short-term reliability, about one hour) and between baseline and follow-up resting recordings (long-term reliability, mean 16-month interval). Correlations (blue) and mean differences (red) are plotted—short term (x-axis) vs. long term (y-axis). Long-term test-retest reliability shows reduced correlations and increased differences. Note that mean differences >0.1 or <−0.1 are typically significant with p < 0.03.

Figure 7

The brain volume for each subject was obtained using Freesurfer (see text). Brain volume for this normative cohort shows the expected decrease with age [23] beginning at about age 60. Over the full age range, the Pearson’s correlation between brain volume and age is −0.295, df = 617, p < 10⁻¹⁸.

Figure 8

Two typical simultaneously active neuroelectric currents were identified and validated by the referee consensus solver, p < 10⁻¹² for each—i.e., p < 10⁻⁴ for each when corrected for multiple comparisons. Each waveform has a duration of 80 msec sampled at 1000 Hz. The bandpass is 10–250 Hz. The currents are 5 mm apart with zero-lag cross-correlation of 0.157, df = 80, p = 0.16. The yellow dot and circle delineate the region near the center of the head which is excluded from the search for neuroelectric currents. See text for details.

Figure 9

The referee consensus solver automatically fails when the recordings are noisy. 300 s of raw magnetoencephalographic (MEG) (lower) and neuroelectric currents (upper) are shown. The number of validated (p < 10⁻¹²) currents identified drops markedly when the MEG is noisy.

Figure 10

The correlations between measures for empty room and human resting for 85 cortical and subcortical regions for the CamCAN (blue, n = 619) and the TEAM-TBI (red, n = 63) cohorts are plotted on the y-axis. The number of events found within each region as a fraction of the total number of events found is plotted on the x-axis; the x-axis is logarithmic. The correlations are greater for TEAM-TBI than CamCAN as expected given the large difference in the n values. See text for details.

Figure 11

Baseline (x-axis) vs. 16-month follow-up (y-axis) z-score regional current densities are plotted for each CamCAN subject (n = 253)—right brain in green, left brain in orange. The left panel shows the hippocampus. Correlations are 0.456 (L) and 0.476 (R); differences are 0.524 (L) and 0.408 (R). The right panel shows the supramarginal cortex. Correlations are 0.524 (L) and 0.482 (R); differences are −0.264 (L) and −0.194 (R). The significance of the differences is readily seen in the displacement of the upward (hippocampus) and downward (supramarginal) distributions. The horizontal and vertical green/orange lines indicate the right/left floor values for the measures. The diagonal lines are 2.0 z-scores from the mean difference. Yellow highlights the areas for which both baseline and follow-up z-scores are ≥2.0.