OUP user menu

Parcellating the neuroanatomical basis of impaired decision-making in traumatic brain injury

Virginia F. J. Newcombe, Joanne G. Outtrim, Doris A. Chatfield, Anne Manktelow, Peter J. Hutchinson, Jonathan P. Coles, Guy B. Williams, Barbara J. Sahakian, David K. Menon
DOI: http://dx.doi.org/10.1093/brain/awq388 759-768 First published online: 9 February 2011


Cognitive dysfunction is a devastating consequence of traumatic brain injury that affects the majority of those who survive with moderate-to-severe injury, and many patients with mild head injury. Disruption of key monoaminergic neurotransmitter systems, such as the dopaminergic system, may play a key role in the widespread cognitive dysfunction seen after traumatic axonal injury. Manifestations of injury to this system may include impaired decision-making and impulsivity. We used the Cambridge Gambling Task to characterize decision-making and risk-taking behaviour, outside of a learning context, in a cohort of 44 patients at least six months post-traumatic brain injury. These patients were found to have broadly intact processing of risk adjustment and probability judgement, and to bet similar amounts to controls. However, a patient preference for consistently early bets indicated a higher level of impulsiveness. These behavioural measures were compared with imaging findings on diffusion tensor magnetic resonance imaging. Performance in specific domains of the Cambridge Gambling Task correlated inversely and specifically with the severity of diffusion tensor imaging abnormalities in regions that have been implicated in these cognitive processes. Thus, impulsivity was associated with increased apparent diffusion coefficient bilaterally in the orbitofrontal gyrus, insula and caudate; abnormal risk adjustment with increased apparent diffusion coefficient in the right thalamus and dorsal striatum and left caudate; and impaired performance on rational choice with increased apparent diffusion coefficient in the bilateral dorsolateral prefrontal cortices, and the superior frontal gyri, right ventrolateral prefrontal cortex, the dorsal and ventral striatum, and left hippocampus. Importantly, performance in specific cognitive domains of the task did not correlate with diffusion tensor imaging abnormalities in areas not implicated in their performance. The ability to dissociate the location and extent of damage with performance on the various task components using diffusion tensor imaging allows important insights into the neuroanatomical basis of impulsivity following traumatic brain injury. The ability to detect such damage in vivo may have important implications for patient management, patient selection for trials, and to help understand complex neurocognitive pathways.

  • traumatic brain injury
  • diffusion tensor imaging
  • decision-making


Cognitive dysfunction is a devastating consequence of traumatic brain injury that affects the majority of those who survive with moderate-to-severe injury, and many patients with mild head injury (Carroll et al., 2004; Frencham et al., 2005). The range of cognitive dysfunction is widespread and includes memory, impaired attention, slowed speed of processing, impaired decision-making, impulsivity, apathy and emotional lability (Bigler, 2008; Wood, 2008). These deficits severely affect the quality of life experienced by patients (Franzen, 2000; Max et al., 2001; Brewer et al., 2002), their interpersonal relationships (Weddell and Leggett, 2006) and their behaviour (including anti-social/criminal aspects) in society at large (Sarapata et al., 1998; Langevin, 2006). In some patients, late imaging shows minimal damage despite obvious impairments in cognition. In others, focal lesions may be present, but may not explain the nature or extent of the cognitive deficits present.

In a previous study from our group, patients have been found to display increased impulsive betting behaviour in a Gambling Task (Salmond et al., 2005b). Such impulsive betting has been associated with abnormalities of the dopaminergic system and may represent one example of impaired decision-making in survivors of traumatic brain injury. Single photon emission computed tomography has also shown reductions in striatal D2 availability post-traumatic brain injury (Donnemiller et al., 2000). While this study was unable to quantify D2 in extra-striatal areas, it does add credence to the hypothesis that dopaminergic pathways may be damaged after traumatic brain injury. Lesions to areas involved in this system, including the ventromedial prefrontal cortex as well as the insular and anterior cingulate cortex, typically cause impulsivity, poor judgement and socially inappropriate behaviour (Clark et al., 2008). In addition, dopaminergic drugs, including levodopa (Debette et al., 2002; Matsuda et al., 2003), amantadine (Van Reekum et al., 1995; Meythaler et al., 2002; Hughes et al., 2005; Sawyer et al., 2008), bromocriptine (Muller and von Cramon, 1994) and methylphenidate (Pavlovskaya et al., 2007), have been reported to improve cognitive function in traumatic brain injury, including the amelioration of impulsivity (methylphenidate) (Pavlovskaya et al., 2007) and reduction in apathy (amantidine, bromocriptine) (Muller and von Cramon, 1994; Van Reekum et al., 1995; Meythaler et al., 2002; Hughes et al., 2005; Sawyer et al., 2008).

Decision-making deficits can be characterized using the Cambridge Gambling Task (Rogers et al., 1999a), a decision-making task that has previously been found to be sensitive to many pathological states including attention-deficit/hyperactivity disorder (DeVito et al., 2008), borderline personality disorder (Bazanis et al., 2002), Huntington’s disease (Watkins et al., 2000) and how gambling preferences change with increasing age (Deakin et al., 2004). This task has been designed to allow decision-making and risk-taking behaviour to be assessed outside a learning context, therefore allowing the neurocognitive deficits in decision-making to be assessed in isolation. In addition, unlike most other ‘Gambling’ tasks, the Cambridge Gambling Task dissociates risk taking from impulsivity, because in the ascending bet condition the subject who wants to make a risky bet has to wait patiently for it to appear. The outcome probabilities of winning and losing are also explicitly provided, thus allowing assessment of decision-making under risk rather than ambiguity.

Some data are available on the neuroanatomical correlates of performance on the Cambridge Gambling Task. Lesion-based studies have implicated frontal lesions in task impairment (Manes et al., 2002; Clark et al., 2003, 2008), and have found functionally dissociable effects of insular and ventromedial prefrontal cortical lesions (Clark et al., 2008). A modified version of the Cambridge Gambling Task, the Risk Task (Rogers et al., 1999b), has been used previously in functional imaging studies. In PET studies of healthy volunteers, significant activations were found in the ventral prefrontal cortex and were predominantly right lateralized (Rogers et al., 1999b; Rubinsztein et al., 2001). Significant activations associated with the resolution of reward conflict have been found in the inferior frontal cortex; the anterior part of the middle frontal cortex, (Brodmann area 10), the orbital gyrus (Brodmann area 11) and the anterior portion of the inferior frontal gyrus (Brodmann area 47) (Rogers et al., 1999b). In a functional MRI task investigating addiction disorders, healthy controls were found to have significant right-sided activation in the lateral orbitofrontal cortex, superior frontal gyrus, dorsolateral prefrontal cortex, inferior temporal gyrus, inferior parietal gyrus, lingual gyrus, occipital cortex and cerebellum (Ersche et al., 2005).

One method that may help to disentangle the complexities of the anatomical basis of such tasks is diffusion tensor imaging. This technique characterizes the diffusion of water molecules in tissue environments that are influenced by the microstructural organization of tissues and their constituent cells. This technique can therefore provide unique insights into pathophysiology of both grey and white matter. The diffusion tensor can be used to represent the magnitude of water diffusion (quantified as the apparent diffusion coefficient), whether such diffusion is directionally non-uniform (anisotropy), and the orientation of that direction (eigenvectors/eigenvalues). These characteristics make diffusion tensor imaging an ideal in vivo tool to assess whether loss of white and/or grey matter integrity may account for some, if not all, of the neuropsychological morbidity post-traumatic brain injury.

Diffusion tensor imaging has been used to study other neurocognitive function in traumatic brain injury with varying success: these include memory and attention (Niogi et al., 2008b), executive function (Nakayama et al., 2006), overall performance on a Mini-Mental State Examination (Nakayama et al., 2006) and learning and memory (Salmond et al., 2006b). Performance in specific cognitive areas (executive function, memory and attention) (Kraus et al., 2007), psychomotor performance (Niogi et al., 2008a), or evoked motor responses (Yasokawa et al., 2007), have been shown to correlate with the overall burden of white matter injury. For example, Sidaros et al. (2007) made the important observation that diffusion tensor imaging abnormalities in specific target regions correlated with broad outcome categories (favourable versus unfavourable on the Glasgow Outcome Scale and the Functional Independence Measure). These results raise the possibility that diffusion tensor imaging measures of traumatic axonal injury may represent markers of the overall severity of traumatic brain injury. Kraus and colleagues (2007) have shown that the extent of traumatic axonal injury abnormalities in multiple brain regions correlates with cognitive performance in specific domains. However, perhaps the best demonstration that localized diffusion tensor imaging abnormalities are directly related to specific cognitive deficits comes from pairwise associations between performance on specific tasks and diffusion tensor imaging measures in cognate brain areas, while demonstrating the absence of correlation with diffusion tensor imaging measures in unrelated brain areas (Niogi et al., 2008b). Thus, in a seminal paper, Niogi et al. (2008b) showed that the diffusion tensor imaging abnormalities in the left corona radiata correlated with attentional performance, while diffusion tensor imaging abnormalities in the uncinate fasciculus correlated with performance on memory tasks. Crucially, the reverse correlations were not shown to be significant. This combination of results (sometimes termed a double dissociation) provides important evidence regarding specificity of regional diffusion tensor imaging abnormalities.

We describe the correlation of diffusion tensor imaging with performance on the Cambridge Gambling Task and use these analyses to characterize the neuroanatomical basis of decision-making deficits following traumatic brain injury.

Materials and methods

Forty-four patients and 40 controls underwent MRI using a 3T Siemens Magnetom Total Imaging Matrix (TIM) Trio. The imaging protocol included diffusion tensor imaging, details of which are described below. Other sequences included a 3D T1-weighted structural sequence (magnetization prepared rapid gradient echo), a fluid attenuated inversion recovery sequence, a gradient echo sequence and a dual spin echo (proton density/T2-weighted) sequence. Patients were imaged at a minimum of 6 months post-injury, and were chosen from a larger cohort, selected because they did not exhibit any significant focal lesions (total lesion volume >1.5 cm3 in size) on fluid attenuated inversion recovery, T2 or gradient echo sequences. In addition, no patients had any focal lesions visible in the regions of interest used, on any of the structural MRI sequences employed. Ethical approval was obtained from the Research Ethics Committee and informed consent was obtained in all cases.

Comprehensive neuropsychological testing designed to test memory, executive function and attention was performed in all patients on the day of imaging. These tasks were part of the Cambridge Neuropsychological Automated Test Battery (CANTAB, www.camcog.com) and were run on an Advantech personal computer (Model PPC-120T-RT), and included the Cambridge Gambling Task. Responses were registered by either a response key or the touch sensitive screen depending on the task, with the subjects using their dominant hand. Eighteen controls (who were not imaged) underwent identical neuropsychological testing. Six months was chosen as the minimum time post-injury, as previous analysis of neuropsychological data in a similar cohort of patients by our group has shown that the neuropsychological parameters assessed by the Cambridge Neuropsychological Automated Test Battery are stable after this time point (Salmond et al., 2006a). Volunteers were recruited from appropriate participants already known to the department and by advertisements placed with ethics committee approval. Exclusion criteria included a prior history of contact with neurological or psychiatric services, previous illicit drug use or alcohol dependence.

To perform the Cambridge Gambling Task, subjects were presented with an array of 10 blue and red boxes, and given a bank of points to bet with. A token was hidden under one of the boxes. The subject was asked to guess under which colour the token was hidden and to wager a proportion of their points on that decision. These wagers were offered in an ascending and descending sequence, in order to differentiate impulsive responses from genuine risk preference, as here the subject must wait to place their bets in the ascend condition. The ratio of blue to red boxes provides the outcome probabilities of winning and losing explicitly, thus allowing the task to assess decision-making under risk rather than ambiguity. The results of the Cambridge Gambling Task were broken down into five components: (i) rational choices, the proportion of trials where the majority colour was chosen; (ii) deliberation time, the latency to make a colour choice; (iii) amount bet, the average across conditions and box ratios (higher bets are assumed to indicate risk preference); (iv) impulsivity index, the difference in percentage bet in descending versus ascending conditions. Consistently early bets (e.g. 95% points descending—5% points ascending) produce a high impulsivity index; and (v) risk adjustment index, quantifies bet calibration across ratios [2 × (% bet 9:1) + (% bet 8:2) − (% bet 7:3) − 2 × (% bet 6:4)]/average % bet, so higher scores imply better risk adjustment (Deakin et al., 2004).

In order to help control for Type 1 errors, previous imaging studies were used to select the regions of interest before commencing the imaging analysis. However these imaging studies (Rogers et al., 1999b; Rubinsztein et al., 2001; Ersche et al., 2005), many of which involved PET, chiefly assessed neuroanatomical involvement of cortical areas, and provided little information on the deep grey matter structures involved in a task. Consequently, the regions of interest specified were expanded by the inclusion of regions based on theoretical knowledge about the circuits proposed to be involved in a particular task. The regions of interest included medial prefrontal cortex, ventrolateral prefrontal cortex, dorsolateral prefrontal cortex, superior frontal gyrus, orbitofrontal gyrus, frontal white matter, hippocampus, insular cortex, thalamus, striatum (dorsal and ventral) and the caudate. In order to detect whether any associations that were detected simply reflected the overall burden of injury, we also sought correlations between task performance in areas that would not be expected to be involved in any aspect of the task, the parietal cortex and the posterior corpus callosum so as to provide negative control regions.

As the majority of regions of interest hypothesized to be involved in the Cambridge Gambling Task contained mainly grey matter, the apparent diffusion coefficient should be more sensitive to the detection of pathological changes than fractional anisotropy. For this reason, and to minimize the number of comparisons, the apparent diffusion coefficient was prospectively chosen as the outcome measure for this study. It has been shown that the use of multiple b-values for a smaller number of unique gradient directions provides apparent diffusion coefficient results that are more robust than ones obtain with a higher number of sampling directions but only one b-value (Correia et al., 2009). Therefore, we used a sequence with multiple b-values. The diffusion tensor imaging parameters were as follows: 12 non-collinear directions, five b-values ranging from 338 to 1588 s/mm2, five b-value = 0 images, acquisition matrix size 96 × 96, field of view 192 mm × 192 mm, 63 axial slices, 2 mm slice thickness, repetition time = 8300 ms, echo time = 98 ms. All scans were visually inspected prior to analysis, and subjects (two patients, four controls) who had moved more than two voxels (4 mm) during the diffusion sequence were removed prior to data analysis. The final dataset was therefore composed of 42 patients and 38 controls.

The diffusion tensor imaging data underwent eddy current correction and apparent diffusion coefficient maps were created using the Oxford Centre for functional MRI of the brain (FMRIB’s) Diffusion Toolbox and all the b-values were used in the calculation of the tensor model (http://www.fmrib.ox.ac.uk/fsl). To aid coregistration, the skull and extracranial soft tissue were stripped from the magnetization prepared gradient echo images using the Brain Extraction Tool (Smith, 2002). The diffusion weighted data were normalized using a two-step approach. First, all patient and control magnetization prepared gradient echo images were coregistered to the MNI152 template using the vtkCISG normalized mutual information algorithm (http://www.image-registration.com). The b = 0 image was subsequently coregistered to the subject’s own magnetization prepared gradient echo image. The transformation matrix normalizing the magnetization prepared gradient echo image was then applied to the b = 0 image.

These regions of interest were manually drawn on the high resolution, high signal-to-noise Colin27 template (Holmes et al., 1998) using Analyse 7.0 (http://www.mayo.edu/bir). This image was chosen as unlike the averaged MNI152 templates, it has enough anatomical detail and contrast necessary to trace regions of interest. All coregistered images were visually inspected to ensure that regions of interest corresponded to the regions specified and/or were not affected by distortion artefact and manually adjusted if they did not. This approach was used to reduce the bias associated with completely hand drawn regions of interest, while mitigating coregistration errors that are inherent in a fully automated approach. The mean apparent diffusion coefficient for the different regions of interest was calculated using in-house software (written by GBW). To ensure the intra-rater reliability of the regions of interest, two regions—the posterior corpus callosum and the thalamus—were completely reanalysed without referring back to the previous adjusted regions of interest and the intra-class correlation coefficients calculated. Apparent diffusion coefficients in each of the regions of interest identified from prior knowledge of task neuroanatomy were compared with performance on the cognate task components.

Statistical analyses were conducted using Statistical Package for the Social Sciences (SPSS 14.0, Chicago, IL, USA, http://www.spss.com). Following assessment of the data for normality, parametric and non-parametric comparisons were performed where appropriate. Mann–Whitney U-test was used for unpaired tests and the Wilcoxon signed rank test for paired comparisons. Partial correlations were used to control both for time to scan post-injury (days) and age at scan (years). Spearman rank correlations were used for non-parametric correlations. To correct for multiple comparisons for the correlations between the regions of interest and the behavioural measures the false discovery rate was calculated which resulted in a P-value of 0.012 (Genovese et al., 2002). For other comparisons P ≤ 0.05 was accepted as significant.



The imaged control group comprised 29 males and nine females, the neuropsychology tested the control group (11 males and seven females) and the patient group (27 males and 15 females) (Table 1). There was no significant difference in mean ± SD age at scan between the head injury group (36.5 ± 14.4 years) and the imaged control group (34.9 ± 10.3) years, P = 0.844) or the neuropsychological testing control group (36.0 ± 11.3 years, P = 0.561). The patient group had a median admission Glasgow Coma Score of 6 (range 3–15) and a median Glasgow Outcome Scale (where higher scores indicate better outcome) of 4 (range 3–5). The mechanisms of injury were as follows: motor vehicle accidents 33 patients, falls in eight patients and alleged assault for one. Patients underwent MRI at a median of 334 (range 171–1342) days.

View this table:
Table 1

Summary of demographic and clinical characteristics of the controls and patients used in this analysis

CharacteristicControls (MRI)Controls (neuropsychology tested)Patients
Age (years)
    Mean ± SD34.9 ± 10.336.0 ± 11.336.5 ± 14.4
    Male, n (%)2911 (61)27 (64)
    Female, n (%)97 (39)15 (36)
    Median, rangeN/AN/A7 (3–15)
Mechanism of Injury
    Road traffic collision, n (%)N/AN/A33 (79)
    Fall, n (%)N/AN/A7 (17)
    Assault, n (%)N/AN/A1 (2)
Marshall gradea
Time from injury to scan, days
    Median, rangeN/AN/A334 (171–1437)
    Median, rangeN/AN/A16.5 (3–36)
    Median, rangeN/AN/A22 (9–50)
Days in ICUa
    Median, rangeN/AN/A10 (1–32)
    Median, IQRN/A29 (28–30)29 (18–30)
    Median, IQRN/A5 (1–10)8 (4–19)
    Median, IQRN/A113 (104–119)110 (102–117)
    Median, rangeN/AN/A4 (3–5)
    Median, rangeN/AN/A6 (3–8)
  • There were no significant differences in any variable between the groups.

  • GCS = post-resuscitation Glasgow Coma Score (Teasdale and Jennett, 1974); Marshall Grade (Marshall et al., 1992) = scored on initial CT. I to IV = diffuse injury, EML = evacuated mass lesion, NEML = non evacuated mass lesion; APACHE = Acute Physiology and Chronic Health Evaluation, higher scores indicate greater severity (Wagner et al., 1984); ISS = Injury Severity Score, higher scores indicate greater severity (Greenspan et al., 1985); MMSE = Mini-Mental State Examination; BDI = Beck Depression Inventory Score at follow-up (higher scores indicate greater severity of depressive symptoms) (Beck, 1970); NART = Estimated pre-morbid intelligence (Nelson, 1982); GOS = Glasgow Outcome Scale (Jennett and Bond, 1975): 1 = death, 2 = persistent vegetative state, 3 = severe disability, 4 = moderate disability, 5 = good recovery; GOSe = Glasgow Outcome Scale Extended: 1 = death, 2 = vegetative state, 3 = lower severe disability, 4 = upper severe disability, 5 = lower moderate disability, 6 = upper moderate disability, 7 = lower good recovery, 8 = upper good recovery.

  • a Eight patients did not require ICU therefore data re ISS/APACHE unavailable.

  • b Missing data due acute computer tomography scan not being available or to patient not completing (Marshall grade n = 1, MMSE n = 1, BDI n = 4, NART n = 5, GOSE n = 5).

Performance on the Cambridge Gambling Task

Compared with controls, the head injured group did not differ in terms of risk adjustment (P = 0.632), rational choices (P = 0.481) or in the amount bet (P = 0.235) (Table 2). The patients exhibited a preference for consistently early bets, as reflected by a significantly higher impulsivity index (P = 0.012). The deliberation time was also significantly slower in the traumatic brain injury group (P = 0.015).

View this table:
Table 2

Summary of results for the Cambridge Gambling Task

Test performedControls (n = 18) (mean ± SD)Patients (n = 42) (mean ± SD)95% Confidence interval of the differenceTest statisticaP-value
Risk adjustment1.59 ± 0.581.49 ± 0.82−0.32 to 0.520.480.632
Impulsivity index42.93 ± 42.8083.74 ± 61.29−72.17 to −9.46−2.640.012
Rational choices1.00 (0.98–1.00)1.00 (0.96–1.00)−0.7050.481
Deliberation time1904 ± 5232484 ± 1087−1044 to −114−2.510.015
Amount bet61.7 ± 9.957.5 ± 11.2−2.77 to
  • Parametric statistics were used for all variables except for rational choices.

  • a Test statistic is the t-score except for rational choices where it is the z-score.

Imaging results and correlation with Cambridge Gambling Task

The intra-class correlation coefficients for intra-rater reliability of mean apparent diffusion coefficient calculated for the posterior corpus callosum and the thalamus were 0.95 and 0.94, respectively. As these intra-class correlation coefficients values are >0.80 they represent excellent agreement of intra-rater reliability.

The mean apparent diffusion coefficient was significantly increased in the patients with post-traumatic brain injury compared with the control group in all regions of interest except the parietal cortex (Table 3). The various task components of the Cambridge Gambling Task correlated significantly with apparent diffusion coefficient in different regions of interest indicating dissociation between components (Table 4, Fig. 1). Increased impulsivity was associated with increasing apparent diffusion coefficient in the orbitofrontal cortices, the insular and the caudate bilaterally. Longer deliberation times were significantly associated with increased apparent diffusion coefficients in the bilateral ventrolateral and dorsolateral prefrontal cortices, superior frontal gyri, orbitofrontal gyri, insular cortices and left sided medial prefrontal cortex. Negative correlations were found with the right thalamus, right dorsal striatum and left caudate for risk adjustment, and the right ventrolateral cortex, bilateral dorsolateral prefrontal cortices and superior frontal gyri, right dorsal striatum right ventral striatum, and left hippocampus for rational choice. No correlations were found with the amount bet.

Figure 1

Neuroanatomical correlates of regional apparent diffusion coefficient in patients with behaviour on Cambridge Gambling Task subcomponents (except for amount bet, which showed no significant correlations). Dissociation between the various components can be clearly seen. (A) Risk adjustment, (B) impulsivity index, (C) rational choices, (D) deliberation time.

View this table:
Table 3

Comparison of mean apparent diffusion coefficients (×10−3 mm2/s) for each region of interest between patients and controls

ControlsTraumatic brain injuryP-valueControlsTraumatic brain injuryP-value
Ventrolateral prefrontal gyrus0.89 ± 0.071.04 ± 0.12<0.0010.87 ± 0.081.03 ± 0.16<0.001
Medial prefrontal gyrus0.92 ± 0.091.07 ± 0.14<0.0010.96 ± 0.061.10 ± 0.13<0.001
Dorsolateral prefrontal gyrus0.84 ± 0.050.94 ± 0.07<0.0010.85 ± 0.060.95 ± 0.11<0.001
Superior frontal gyrus0.83 ± 0.040.91 ± 0.030.0020.82 ± 0.030.93 ± 0.02<0.001
Orbitofrontal gyrus0.77 ± 0.100.88 ± 0.140.0010.79 ± 0.090.88 ± 0.100.001
Frontal white matter0.63 ± 0.020.74 ± 0.09<0.0010.65 ± 0.020.75 ± 0.10<0.001
Hippocampus0.75 ± 0.080.81 ± 0.03<0.0010.75 ± 0.040.83 ± 0.07<0.001
Insular cortex0.76 ± 0.070.88 ± 0.11<0.0010.75 ± 0.090.81 ± 0.08<0.001
Thalamus0.73 ± 0.050.88 ± 0.01<0.0010.70 ± 0.070.81 ± 0.06<0.001
Dorsal striatum0.88 ± 0.121.06 ± 0.21<0.0010.90 ± 0.121.09 ± 0.18<0.001
Ventral striatum0.76 ± 0.070.79 ± 0.06<0.0010.74 ± 0.070.81 ± 0.06<0.001
Caudate0.90 ± 0.021.10 ± 0.03<0.0010.94 ± 0.031.23 ± 0.03<0.001
Parietal cortex0.92 ± 0.070.93 ± 0.070.2150.92 ± 0.050.94 ± 0.060.568
Posterior corpus callosum0.88 ± 0.141.13 ± 0.03<0.0010.89 ± 0.101.14 ± 0.03<0.001
  • Mean values ± standard deviations are shown.

View this table:
Table 4

Correlation coefficients for mean apparent diffusion coefficient values for each region of interest with components of the Cambridge Gambling Task in patients at least 6 months post-traumatic brain injury

Risk adjustmentP-valueImpulsivity indexP-valueRational choiceP-valueAmount betP-valueDeliberation timeP-value
    Medial prefrontal cortex0.1710.7920.1370.448−0.2200.238−0.3140.0750.2690.162
    Ventrolateral prefrontal cortex−0.0880.6270.2720.1260.3420.000−0.0940.6030.5880.000
    Dorsolateral prefrontal cortex−0.1380.8780.1100.5410.3120.004−0.0400.8250.5900.000
    Superior frontal gyrus−0.2550.1530.1420.9920.3130.0030.0300.6330.4660.006
    Orbitofrontal gyrus−0.0990.2370.6740.000−0.3170.063−0.2290.8350.6140.000
    Frontal white matter−0.1520.4670.2820.112−0.1400.4220.1810.8990.2490.098
    Insular cortex−0.1520.7430.6200.000−0.1260.254−0.2080.4800.5050.000
    Dorsal striatum0.4810.0050.1100.2820.3860.0010.1820.1730.3350.027
    Ventral striatum−0.1490.6270.3350.0570.2860.0030.0790.6600.2930.098
    Parietal cortex−0.0540.764−0.2380.356−0.1630.180−0.0550.1160.2380.667
    Posterior corpus callosum−0.1620.4900.0850.828−0.1210.6720.0910.7440.1010.347
    Medial prefrontal cortex−0.1060.8920.0350.845−0.2110.0220.1120.5360.5550.001
    Ventrolateral prefrontal cortex0.0370.8390.1660.355−0.3360.019−0.1340.4580.5280.002
    Dorsolateral prefrontal cortex−0.0280.4440.0080.9630.3870.004−0.0800.6590.4380.001
    Superior frontal gyrus−0.3000.0890.0020.4310.3540.007−0.0050.7120.5650.001
    Orbitofrontal gyrus−0.0690.1230.6830.000−0.3290.053−0.1690.8250.6600.000
    Frontal white matter−0.2670.1660.0270.882−0.0340.8450.0250.7390.1790.452
    Insular cortex−0.1520.2650.5670.001−0.2630.112−0.3110.3570.5820.000
    Dorsal striatum−0.4280.0250.0930.608−0.2850.0220.2430.3100.1690.248
    Ventral striatum−0.2810.4080.0390.828−0.1890.1410.0980.5860.0300.868
    Parietal cortex−0.0290.871−0.1500.426−0.1250.1070.1060.2480.1870.563
    Posterior corpus callosum−0.1240.1220.0390.156−0.1400.4300.1690.6620.0590.287
  • The figures in the table are partial correlation coefficients, controlling for time to scan since injury and age at scan, for parametric statistical analyses, and Spearman’s rho (ρ) for non-parametric statistical analyses (which was only used for rational choices). The bold values are regions of interest that had significant correlations with P ≤ 0.012 (false discovery rate corrected).


Patients who had sustained traumatic brain injury were found to have broadly intact processing of risk adjustment and probability judgement, and bet similar amounts to controls, indicating a similar level of risk adjustment to the controls. The patient preference for consistently early bets indicated a higher level of impulsiveness, which may be secondary to an intolerance of delay to rewards. While these group level analyses provide some data reading the anatomical basis of impaired decision-making following traumatic brain injury, the ability to correlate the magnitude of imaging abnormality with task performance provided important additional insights.

This was a group of traumatic brain injury survivors who exhibited a range of impaired performance on the Cambridge Gambling Task, and the ability to associate the location and extent of damage with performance on the various task components allows important insights into the neuroanatomical basis of performance on such tasks. Impulsivity appeared to be associated with apparent diffusion coefficient in the orbitofrontal gyrus, insular and caudate, consistent with studies of impulsivity in attention deficit hyperactivity disorder (Volkow et al., 2007; Brennan and Arnsten, 2008), and risky decision-making in healthy volunteers (Xue et al., 2010). Abnormal risk adjustment correlated with apparent diffusion coefficient in the right thalamus, right dorsal striatum and left caudate, and impaired performance on rational choice was significantly correlated with apparent diffusion coefficient in the ventro- and dorsolateral prefrontal cortices, the superior frontal gyrus, the right dorsal and ventral striatum and left hippocampus. Further, the absence of associations between task performance and imaging abnormality in unrelated brain regions, attests to the specificity of our findings. These data thus provide a more complete picture of how traumatic brain injury impacts on the complex circuitry involved in decision-making.

This specificity and sensitivity of diffusion tensor imaging contrasts with unimpressive results in previous studies of decision-making in survivors of traumatic brain injury, which have focused on more conventional structural imaging. In a large study of 71 traumatic brain injury patients by Levine and colleagues (2005), significant deficits were common on the Iowa Gambling Task (Bechara et al., 1994), but bore no relation to lesion size, lesion location or amount of atrophy detected on T1, T2 or gradient echo sequences. The lack of correlation with conventional structural MRI was confirmed in a subsequent study by this group, which found that the Iowa Gambling Task was not consistently related to the structural integrity of the ventral frontal cortex in patients with lesions or traumatic axonal injury (Fujiwara et al., 2008). Only in the focal lesion group was any significant correlation found, which was confined to the grey matter volumes in the superior frontal cortex (Fujiwara et al., 2008). Neither study found a relationship between injury severity and task performance (Levine et al., 2005; Fujiwara et al., 2008).

The finding of correlations with grey matter injury and neurocognitive outcome in patients with traumatic axonal injury is not surprising. The majority of diffusion tensor imaging studies performed in traumatic brain injury have concentrated on white matter changes. However, the technique may also illustrate grey matter injury as shown here. This injury may occur not only as a consequence of direct trauma, as is the case of contusions, but also secondary to the effects of traumatic axonal injury. Damage to the grey/white matter junction, oxidative metabolic dysfunction (Xu et al., 2010) and Wallerian degeneration may all lead to neuronal atrophy, loss and/or dysfunction. A reduction in grey matter volume in chronic traumatic brain injury has been found in many brain regions including the frontal and temporal cortices, subcortical grey matter and the cerebellum (Gale et al., 2005; Salmond et al., 2005a). Neuronal loss using flumazenil PET has also been noted in brain regions without other MRI abnormalities (Shiga et al., 2006).

One striking feature of our neuropsychological data was the psychomotor slowing of patients with traumatic brain injury. Such slowing is a common manifestation of traumatic brain injury, occurring in both the perceptual interpretation of stimuli and in motor responses (Lew et al., 2009). This generalized slowing of processing ability correlated with apparent diffusion coefficient in many areas of the brain, particularly the frontal cortices. This is in agreement with previous work by Niogi and colleagues (2008a), which found the number of damaged white matter structures as quantified by diffusion tensor imaging significantly correlated with mean reaction time.

The use of the Cambridge Gambling Task may have particular advantages in assessing the quality of decision-making in survivors of traumatic brain injury who often have memory impairment. In a previous study from our group using the Cambridge Gambling Task, patients with traumatic brain injury were found to be slower to make choices, were more impulsive (consistent with orbitofrontal cortex damage), and poorer at decision-making than controls (Salmond et al., 2005b). Most other traumatic brain injury studies assessing decision-making have used the Iowa Gambling Task (Bechara et al., 1994, 2000). This task is thought to be particularly sensitive to ventral medial orbitofrontal cortex and/or amygdala lesions (Levine et al., 2005; Hanten et al., 2006; Garcia-Molina et al., 2007; Bonatti et al., 2008; Fujiwara et al., 2008). In this task, the patient learns to associate reward and punishment with four card decks while earning pretend money. Unlike the Cambridge Gambling Task, this task therefore is a ‘risk anticipation task’ testing decision-making under conditions of ambiguity. It therefore may involve emotional processing, working memory and attention, all of which may be impaired in patients post-traumatic brain injury. Indeed, survivors of traumatic brain injury who are able to learn the task’s reinforcement contingencies have been found to perform better than those who do not (Garcia-Molina et al., 2007), and in a study of 11 children post-traumatic brain injury, memory impairment was related to performance on a modified Iowa Gambling Task (Hanten et al., 2006). Recent Iowa Gambling Task data have questioned the specificity of the ventromedial cortex in causing decision-making deficits in patients post-traumatic brain injury (MacPherson et al., 2009). Patient performance was equally impaired in those with and those without ventromedial cortical lesions (MacPherson et al., 2009). The lack of correlation in our patients with this area in any task component is consistent with these results. It should also be noted that the majority of patients in both groups in this study had lesions identifiable in the dorsolateral prefrontal cortex, an area that we found to correlate with rational choice and deliberation time (MacPherson et al., 2009).

There are numerous reasons for cautious interpretation of the data presented here. Although all images were visually inspected and manually corrected where needed, it is possible that residual coregistration errors may persist. The correlation of structural integrity with behavioural outcomes is also complicated by issues of functional compensation and neuronal plasticity (Johansen-Berg, 2007). Despite these caveats, it is telling that although none of the patients included in this study had large focal lesions, regions that were hypothesized to be correlate with impaired test performance (based on prior imaging data in normal subjects) did show such correlations.

This study therefore adds to the evidence that microstructural integrity, as detected by diffusion tensor imaging, is an important determinant of function post-traumatic brain injury. This compromised microstructural integrity leads to slow deliberation times and impulsive responding. This may lead to problems and failures in decision-making in daily life, particularly in time-limited situations, consequently impacting on the patients’ quality of life and wellbeing. The ability to detect such damage in vivo may have important implications for patient management, aid selection and stratification of patients for clinical trials, and help understand complex neurocognitive pathways.


Medical Research Council (UK) Program Grant [Acute brain injury: heterogeneity of mechanisms, therapeutic targets and outcome effects (G9439390 ID 65883)]; UK National Institutes of Health Research Biomedical Research Centre at Cambridge; UK Department of Health, Technology Platform funding. Gates Cambridge Trust (to V.F.J.N.); Overseas Research Studentship (to V.F.J.N.); National Institute for Health Research Academic Clinical Fellow (to V.F.J.N.); Academy of Medical Sciences/Health Foundation Senior Surgical Scientist Fellowship (to P.J.H.); Academy of Medical Sciences/Health Foundation, Clinician Scientist Award (to J.P.C.); Wellcome Trust Programme Grant (No 089589/Z/09/Z to B.J.S.); BOC Professorship of the Royal College of Anaesthetists (to D.K.M.); Evelyn Trust (to D.K.M.).


The authors thank Dr S. Sawiak from the Wolfson Brain Imaging Centre for his statistical advice.


View Abstract