Abstract
Objective: To assess patterns of regional brain activation in response to varying working memory loads shortly after mild traumatic brain injury (MTBI).
Background: Many individuals complain of memory difficulty shortly after MTBI. Memory performance in these individuals can be normal despite these complaints.
Methods: Brain activation patterns in response to a working memory task (auditory n-back) were assessed with functional MRI in 12 MTBI patients within 1 month of their injury and in 11 healthy control subjects.
Results: Brain activation patterns differed between MTBI patients and control subjects in response to increasing working memory processing loads. Maximum intensity projections of statistical parametric maps in control subjects showed bifrontal and biparietal activation in response to a low processing load, with little additional increase in activation associated with the high load task. MTBI patients showed some activation during the low processing load task but significantly increased activation during the high load condition, particularly in the right parietal and right dorsolateral frontal regions. Task performance did not differ significantly between groups.
Conclusion: MTBI patients differed from control subjects in activation pattern of working memory circuitry in response to different processing loads, despite similar task performance. This suggests that injury-related changes in ability to activate or to modulate working memory processing resources may underlie some of the memory complaints after MTBI.
The majority of the almost 2 million traumatic brain injuries sustained each year in this country1 can be considered mild by conventional criteria.2 From a cognitive standpoint, most studies suggest that deficits in attention, working memory, and speed of information processing can be found in mild traumatic brain injury (MTBI) patients in the first several weeks after the injury.3,4 Due to the subtlety of the deficits, individuals with MTBI may perform normally on cognitive tests unless they are placed under mild physiologic stress.5 Furthermore, it is common to see MTBI patients who have normal CT or MRI scans of the brain, yet have significant postconcussive complaints and measurable cognitive deficits. PET and SPECT have been reported to show regional abnormalities in TBI patients even when CT and MRI scans are normal.6,7
Compared with PET and SPECT, noninvasive functional MRI (fMRI) has several advantages, including improved spatial and temporal resolution and absence of exposure to radiation, allowing for multiple scans. The technique capitalizes on the different magnetic properties of oxyhemoglobin and deoxyhemoglobin. Their ratio is altered in response to changes in local blood flow associated with increased regional neuronal activity. This change in ratio, or blood oxygen-level-dependent contrast can be used as an indirect measure of changes in blood flow and local metabolic activity. Several studies have used fMRI techniques to outline the brain regions associated with working memory,8,9 which involves the “online” storage of information necessary for performing cognitive operations.10,11 A key feature of working memory is the amount of information that must be held online to solve a particular problem (“working memory load”). Most studies of brain regions involved in working memory have been conducted with normal individuals and have found bilateral frontal and parietal activation, with increased working memory load associated with specific increases in activation.12 We are unaware of any systematic fMRI studies of individuals with TBI.
We hypothesized that individuals with MTBI have a physiologic basis to their memory and attentional complaints, and that fMRI performed within 1 month of their injury would show changes in activation of working memory circuitry in response to changes in processing load.
Methods.
This was a prospective, controlled study of consecutive patients with MTBI referred to the Dartmouth Hitchcock Medical Center (DHMC). DHMC is a level 1 trauma center and is the major referral center for a 150-km radius in northern New England. All participants gave informed consent, as approved by the Dartmouth Medical School Committee for the Protection of Human Subjects.
Subjects.
Individuals were considered to have sustained an MTBI if they met the criteria established by the American Congress of Rehabilitative Medicine for mild brain injury.13 Specifically, they had sustained a traumatic blow to the head, resulting in either alteration of level of consciousness (manifested by being dazed and confused and having amnesia for the event) or loss of consciousness (LOC) of less than 30 minutes, a period of post-traumatic amnesia (PTA) less than 24 hours, and a Glasgow Coma Scale (GCS) score of 13 to 15 on arrival at the DHMC emergency department. Duration of LOC was estimated by using all available information including patient and witness reports, emergency personnel records, and DHMC medical records. PTA was estimated by careful questioning of patients to determine the time of return of continuous memory as well as by review of medical records. Five patients were involved in motor vehicle crashes, four were injured in falls, and three had injuries related to sports and recreation.
MTBI patients were excluded if they had a history of other neurologic disorders (such as epilepsy, cerebrovascular disease, mental retardation, neurodegenerative disorders, prior TBI with clear LOC), significant systemic medical illness, or current Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV),14 axis I diagnosis of psychiatric illness other than substance abuse. The Structured Clinical Interview for DSM-IV15 was used to screen for psychiatric illness. During the period of recruitment for this study, the records of 56 potential participants were screened for study inclusion. Of those not included in this study, 44% did not participate because they lived out of state or at a great distance; 44% because they had medical, neurologic, or psychiatric diagnoses; and 12% did not respond to multiple attempts to contact them.
Healthy control subjects were recruited through advertisements in local newspapers and workplaces, and were screened for neurologic, medical, or any past or current psychiatric illness.
Clinical assessment.
Following resolution of PTA, patients underwent a comprehensive screen of medical and psychiatric history as well as questions regarding their current symptoms. The head injury symptom checklist used to evaluate current symptoms was adapted from both Dikmen et al.16 and the Philadelphia Head Injury Questionnaire.17 Only items not present before the trauma, or increased notably in frequency or intensity since the trauma, were scored as related to the brain injury. An 18-item memory self-rating scale18 was used to assess subjective perception of memory function. Participants were given a battery of neuropsychological tests designed to assess level of general intellectual function, speech and language function, attention/concentration, memory, and executive function.
Working memory tasks.
We used the auditory “n-back” task to probe working memory. Our auditory version was modeled after tasks used in a number of previous studies investigating working memory in normal control subjects.19 During scanning, participants were asked to listen to a string of consonant letters (except L, W, and Y) presented every 3 seconds. Three conditions were presented: 0-back; 1-back; and 2-back. The 0-back control condition had a minimal working memory load; individuals were asked to decide whether the current letter matched a single target letter that was specified before the epoch began. During the 1-back condition they were asked to decide whether the current letter matched the previous one. During the 2-back condition the task was to decide whether the letter currently presented matched the letter that had been presented two back in the sequence. On hearing a match, participants were asked to squeeze a pneumatic response bulb to the correct targets. The number of correct and incorrect responses was recorded. The scanning run lasted 288 seconds, during which time whole-brain volumes (20 to 23 slices) were acquired every 3 seconds. There were 12 epochs during the scanning run. The 1-back and 2-back conditions were presented in 24-second epochs, whereas the 0-back control condition lasted 18 seconds. Each epoch was preceded by 3 seconds of instruction (e.g., “squeeze the bulb when you hear an H” or “squeeze when the letter matches one/two back”). The 0-back control condition appeared six times and always alternated with the two experimental conditions. The experimental conditions were presented three times each. The order of presentation is summarized as follows: 0, 1, 0, 2, 0, 1, 0, 2, 0, 1, 0, 2. During each epoch there was a possibility of one, two, or three matches, and the number of matches was counterbalanced within and across conditions. Participants rehearsed the tasks outside the scanner to ensure understanding of task demands.
Functional MRI scan procedure.
Scans were acquired using a 1.5-T GE Signa scanner (Milwaukee, WI) and a multiaxial local gradient head coil system (Medical Advances, Inc., Milwaukee, WI). A gradient echo echo-planar sequence was designed to provide whole-brain coverage: repetition time, 3,000 msec; echo time, 40 msec; field of view, 24 cm (20 or 23 6-mm-thick sagittal slices with no skip); and a 64 × 64 matrix with 3.75-mm in-plane resolution. Before scanning, linear shims were optimized. A time series of 98 T2*-weighted volumes was acquired during the task. The first two volumes were discarded because spins were not saturated completely.
Preprocessing procedures.
All scans were cropped to eliminate most nonbrain voxels. Spatial realignment using the SPM96 six-parameter model (Wellcome Department of Cognitive Neurology, University College, London) was performed on all raw scan data before further analysis to remove any minor (subvoxel) motion-related signal change. Before multisubject analyses, mean images were normalized spatially to the Montreal standardized atlas space using a 12-parameter affine approach and a T2*-weighted template image. The optional use of nonlinear warping by spatial basis functions was limited to 2 × 2 × 2 and eight iterations. During spatial normalization all scans were resampled to 2-mm3 isotropic voxels. Spatial smoothing to a full width half maximum of 15 mm3 was then performed.
Statistical analysis.
fMRI analyses included statistical parametric mapping, on a voxel by voxel basis, using a general linear model approach20 as implemented in SPM96. The task was analyzed initially for each participant as an individual time series before inclusion in multisubject and covariance analyses. The main analyses reported here used the random effects procedure recently developed by Holmes and Friston.21 The principal advantage of this method is the elimination of highly discrepant variances between and within individuals in constructing an appropriate error term for hypothesis testing and generalizability to the population. For the multisubject/between-group analyses, the random effects procedure assumes input of one scan per individual for each condition and then performs a mixed model analysis to account for both random effects (scan) and fixed effects (task conditions). The mean input images for each individual were obtained by calculating the mean image for the 0-back, 1-back, and 2-back conditions for each participant after offsetting by 3 seconds (one repetition time) to account for the hemodynamic response function. The scans during which instructions were administered were discarded. A general linear model analysis was performed on a voxel-by-voxel basis. A parametric design was implemented to test the hypothesis of between-group activation differences as a function of working memory processing load.
For a priori hypothesis testing, critical probability thresholds were uncorrected and set at 0.001 for assessing predefined search regions, including the prefrontal and parietal cortices and interconnected components of the attentional network activated in PET and fMRI studies.9,22 In view of our neuroanatomically constrained hypotheses regarding expected regions of activation based on prior functional imaging and lesion studies, a multiple-comparison correction strategy designed for exploratory searches of the entire brain volume would have been overly conservative.
Results.
Twelve MTBI patients and 11 control subjects were studied. MTBI patients were studied a mean of 22.1 days after their injury (SD, 10.5 days; range, 6 to 35 days). All patients and control subjects were right-handed. Table 1 describes the basic demographic characteristics of both groups. The groups did not differ significantly with respect to gender, age, affective status, level of anxiety, estimates of preinjury general intellectual function, years of education, or years of parental education. All MTBI patients had GCS scores of 15 when seen in the DHMC emergency department except one patient, who had a score of 14. Duration of LOC ranged from 1 to 30 minutes. PTA ranged from 15 minutes to 24 hours. All CT scans and structural MR images were read as normal except for one individual who had a basilar skull fracture.
Demographic characteristics of the mild traumatic brain injury and control groups
MTBI patients endorsed significantly more cognitive symptoms on the head injury symptom checklist than did the control subjects (p < 0.004). In particular, they endorsed items indicating poor memory for recent events, trouble concentrating, and difficulties doing their job. MTBI patients reported significantly more problems with memory on the memory self-rating scale. They rated their memory as worse than control subjects on all 18 items, eight of which achieved statistical significance (p < 0.05).
Despite these concerns regarding their cognitive function, MTBI patients performed generally as well as control subjects on the neuropsychological battery, differing only in response speed on both the simple reaction time and distractibility tasks of the Continuous Performance Test (table 2). There was no statistically significant difference in performance on the three n-back tasks between the MTBI patients and the control subjects.
Neuropsychological performance data for the mild traumatic brain injury and control groups
Functional MRI results.
Analysis of fMRI results showed several foci of activation associated with increasing working memory processing load. Figure 1 shows the maximum intensity projections of the statistical parametric maps of n-back task comparisons for both MTBI patients and control subjects. Figure 2 shows major activation foci in a surface-rendered projection displayed on a standardized atlas brain. As can be seen in both figures 1 and 2⇓ (display threshold, p < 0.01; extent ≥ 172 voxels), both groups showed significant bilateral frontal and bilateral parietal activation in response to increases in working memory processing load.
Figure 1. Maximum intensity projections of the statistical parametric maps of n-back task comparisons for patients with mild traumatic brain injury (MTBI; B) and control subjects (A). The top panel shows areas of significant increase from the 0-back to 1-back condition (left) and 1-back to 2-back condition (right) in control subjects. The lower panel shows results of the same analysis for MTBI patients. Both groups showed significant bifrontal and bilateral parietal activation, although the pattern was different in response to increases in working memory load. The most striking difference is seen in the 1-back to 2-back comparison, in which control subjects required only a focal right frontal increase but MTBI patients required widespread continued activation in working memory regions.
Figure 2. The location of major cortical activation foci are displayed on a surface-rendered projection. The gyral location of activations (bilateral dorsolateral prefrontal and superior parietal) were similar in both groups from the 0-back to 1-back condition. Major differences, as described in the text, were observed in the 1-back to 2-back comparison. Note the more extensive activation of primarily right superior parietal and dorsolateral prefrontal cortex in patients with mild traumatic brain injury (MTBI).
The hypothesis of working memory load-related differences between MTBI patients and control subjects was confirmed. As shown in figures 1 and 2⇑, control subjects showed an increase in regions involved in working memory from the 0-back to the 1-back condition with minimal additional increases from the 1-back to the 2-back condition. By contrast, MTBI patients showed significant but less activation than control subjects in the 0-back to 1-back comparison, but showed extensive activation as working memory load increased from 1-back to 2-back. Also noteworthy is the highly focal nature of the increase from 1-back to 2-back in the control subjects compared with the extensive frontal and parietal increases seen in the MTBI group. Peak amplitude values are similar across regions. Table 3 provides Talairach coordinates, region definitions, and amplitude for the main effect of processing load (0-back, 1-back, and 2-back) across patients and control subjects, as well as the simple effects for each group. An additional SPM96 analysis designed to indicate regions differing as a function of processing load (2 > 1) yielded two major foci (right frontal: Z = 2.79, p = 0.003, x, y, z = 28, 32, 34 mm; right parietal: Z = 2.43, p = 0.008, x, y, z = 24, −78, 42 mm). Activation results for all participants in regions showing significant group-by-working memory load interactions are shown in figure 3. The figure shows adjusted signal change from the 0-back to the 1-back conditions and the 1-back to the 2-back conditions for control subjects and MTBI patients in the right prefrontal cortex (top) and the right parietal cortex (bottom). A similar interaction pattern was shown in both regions, with control subjects primarily showing an increase from 0-back to 1-back and patients showing primarily an increase from 1-back to 2-back.
Talairach coordinates, region definitions, and amplitude for main effect and simple effects of processing load and group
Figure 3. Activation results for all participants in regions show significant group-by-working memory load interactions. The panel shows adjusted signal change from the 0-back to the 1-back conditions, and from the 1-back to the 2-back conditions for control subjects and patients with mild traumatic brain injury in the right prefrontal cortex (top) and the right parietal cortex (bottom). A similar interaction pattern was shown in both regions, with control subjects showing primarily an increase from 0-back to 1-back and patients showing primarily an increase from 1-back to 2-back.
To examine the relationship between activation and performance on the n-back as well as other neuropsychological tests related to attention and memory, we extracted the covariate-adjusted data for all participants at all processing load conditions for the activation foci shown in figures 1 and 2⇑. The Pearson Product Moment correlation coefficients indicated a significant positive relationship between left frontal regions of interest (ROIs) and mean accuracy on the n-back activation task. The highest correlation was observed between n-back mean accuracy and the left lateral frontal ROI (−30,12,54 mm; r = 0.66, p < 0.001). Accuracy on the Continuous Performance Test distractibility subtest, administered as part of the neuropsychological test battery, showed a nearly identical pattern of correlations (r = 0.67, p < 0.001 for left lateral frontal ROI).
Discussion.
Our results indicate that the auditory n-back task activates a regional network similar to that seen with other forms of working memory tasks in normal control subjects23 including areas of frontal and parietal cortex. Activation of this network was observed in both the 1-back and 2-back tasks relative to the 0-back condition. It has been suggested that there is, in general, a linear relationship between working memory load and increase in activation,23 however we did not find such a simple relationship with the auditory n-back task. Control subjects appeared to activate their processing resources primarily when moving from simple vigilance (0-back) to a low-demand working memory task (1-back). These differences may be accounted for in part by the modality of presentation of the working memory tasks. Most research groups have used visual presentation of n-back stimuli, whereas we employed an auditory version.
It is important to note that when moving to the tasks with increased processing demands control subjects showed no significant decline in performance, suggesting that the differences in regional brain activation relate more to working memory processing load rather than to task difficulty. If these changes were due to a nonspecific effect of task difficulty alone, one might expect to see a large increase in brain activation between the 1-back and 2-back conditions, and concomitant declines in performance. The absence of a significant decline in performance as well as a lack of dramatic increase in activation between the 1-back and 2-back tasks in control subjects argues against this notion.
Patients with MTBI studied within 1 month of their injury showed activation of a regional network similar to that seen in the control subjects. As processing load increased, both MTBI patients and control subjects activated more brain area but maintained a similar overall topographic pattern of activation, and similar levels of task performance. The overall magnitude of change in activation in frontal and parietal cortical areas when comparing the 0-back control task with the 2-back task was similar between the two groups. However, the interaction of working memory processing load and degree of increased brain activation differed between the two groups, particularly in the right dorsolateral frontal and right lateral parietal regions. In those areas, the control subjects showed increased activation going from the 0-back to the 1-back condition, and a much smaller increase going from the 1-back to the 2-back condition. By contrast, the MTBI patients showed relatively little increase going from the 0-back to the 1-back condition, but significantly more activation when going from the 1-back to the 2-back condition.
The explanation for this difference in activation pattern between patients and control subjects is not immediately obvious. It is unlikely that these activation differences can be attributed to either attentional differences between the groups or to level of affective distress, such as anxiety or depression, because the groups performed similarly on cognitive tasks assessing attention and neither group endorsed significant anxiety or depressive symptoms (see tables 1 and 2⇑). Models of verbal working memory have proposed at least three different components9,23 including a rehearsal component (postulated to be in frontal speech areas such as Broca’s, and supplemental motor areas), an attentional component (postulated to be in the posterior parietal area), and a processing or executive component, which allows for the manipulation and modification of information (usually considered to be in the prefrontal area). The regions of task-associated activation in both our control and MTBI groups include the areas subserving these three components of verbal working memory and are consistent with the work of others.8,9 The frontal cortex is often injured in TBI, and thus it would not be surprising to find problems with working memory circuitry after sustaining this type of injury. Given the very mild degree of injury experienced by our MTBI group, we would not expect that these individuals exhibited widespread neuronal loss. The fact that our MTBI group ultimately showed similar degrees of activation in response to the processing demands of the 2-back task is consistent with the notion that the group differences are not simply a result of neuronal loss from MTBI. Rather, the ability to activate, modulate, or allocate processing resources in response to gradations of processing load may be impaired in the postacute period after MTBI.
There were relatively few differences in performance found between the two groups on either the n-back tasks or the neuropsychological battery. However, despite equivalent performance, the MTBI patients reported significantly more symptoms, particularly difficulties with memory. One possible explanation is that the MTBI patients perceive the change in their ability to engage working memory easily and efficiently, and experience this change as “having to work harder” to maintain accurate task performance. Perhaps this is then labeled as “problems with memory.” If true, this might account for the discrepancy between the severity of complaints voiced by many MTBI patients and the relatively minor performance deficits often found in these individuals.
It is interesting to note the activation pattern seen in control subjects. The near-maximal activation seen when going from the 0-back condition, which involves primarily vigilance, to the 1-back task was not predicted. We were expecting a more linear relationship between processing load and degree, and spatial extent of activation.19 The observed pattern suggests that activation of the working memory network may be more complicated than a direct load to degree of activation relationship. Perhaps even simple processing loads require that the entire system be “turned on.” Modest increases in processing load (e.g., going from the 1-back to the 2-back task) can be handled by the normally functioning network once it has been activated. The smaller level of activation seen in the MTBI group when moving from the 0-back to 1-back condition could represent a problem in switching on the network. If perceived by the MTBI group, this could be labeled as memory problems.
Alternatively, this pattern might be explained by differences in anticipation of processing load. All participants (control subjects and MTBI) were instructed carefully in the n-back tasks and rehearsed them before going into the scanner. Perhaps there were group differences in ability to anticipate processing load requirements. Control subjects were able to anticipate the degree of task difficulty associated with the maximum processing load, and activated their working memory circuit in anticipation of future processing load requirements. MTBI patients may have responded to immediate task difficulty only.
In any case, given the similarity in topographic pattern, overall degree of activation to moderate processing load, and task performance between the two groups, the differences seen between control subjects and this group of very mild TBI patients appear to have more to do with the timing, allocation, and modulation of processing resources than an actual decrement in available resources.
These interpretations should be tempered by several issues. Although the MTBI patients were consecutive admissions to the DHMC emergency department who agreed to participate in the study, not all MTBI patients agreed to participate, and our admission criteria ruled out those with marked psychopathology. Thus these individuals may not be representative of all MTBI patients. The working memory task reported here, although representative of this cognitive function, is of course not representative of other cognitive functions. However, given that working memory complaints are among the most common after MTBI, it seemed an important area to investigate. It will also be helpful to explore the stability of the fMRI measures. MTBI patients and control subjects are currently being restudied 1 year after their original evaluation and will be the subject of a separate report. We hypothesize that MTBI patients will show activation patterns more similar to those of normal control subjects 1 year after their injury, when the majority will likely show complete recovery.
This needs to be explored further with increasing working memory processing loads that might stress the system more fully. We are currently engaged in such an experiment using tasks with a higher memory processing load.
Whatever the explanation, the discrepancy between MTBI patient complaints and actual task performance is at the heart of the controversy about the etiology of the postconcussive syndrome, and frequently results in attributing MTBI patient complaints to malingering or psychopathology such as depression, anxiety, post-traumatic stress disorder, or other mechanisms. This is common given the typically normal structural imaging results. This study describes significant differences in patterns of brain activation in response to working memory tasks, and suggests that an alteration in the ability to activate or to allocate processing resources in response to a moderate working memory task may be associated with cognitive complaints after very mild TBI. These findings are also consistent with other recent data from our laboratory24 suggesting fMRI has potential clinical applications in cognitive disorders and that the patterns of activation differences are likely to be more important than the overall level of activation. fMRI seems likely to evolve as a helpful tool for the investigation of sequelae of MTBI.
Acknowledgments
Supported in part by National Institute on Disability and Rehabilitation Research grant no. H133G70031 and the Ira DeCamp Foundation.
Acknowledgment
The authors would like to thank Chad Moritz and Robert Ferranti for their assistance with scanning.
- Received December 31, 1998.
- Accepted April 29, 1999.
References
Disputes & Debates: Rapid online correspondence
REQUIREMENTS
If you are uploading a letter concerning an article:
You must have updated your disclosures within six months: http://submit.neurology.org
Your co-authors must send a completed Publishing Agreement Form to Neurology Staff (not necessary for the lead/corresponding author as the form below will suffice) before you upload your comment.
If you are responding to a comment that was written about an article you originally authored:
You (and co-authors) do not need to fill out forms or check disclosures as author forms are still valid
and apply to letter.
Submission specifications:
- Submissions must be < 200 words with < 5 references. Reference 1 must be the article on which you are commenting.
- Submissions should not have more than 5 authors. (Exception: original author replies can include all original authors of the article)
- Submit only on articles published within 6 months of issue date.
- Do not be redundant. Read any comments already posted on the article prior to submission.
- Submitted comments are subject to editing and editor review prior to posting.