J Neurol Neurosurg Psychiatry 83:1193-1200 doi:10.1136/jnnp-2012-302644
  • Neurosurgery
  • Research paper

A longitudinal MRI study of traumatic axonal injury in patients with moderate and severe traumatic brain injury

  1. Anne Vik1,2
  1. 1Department of Neuroscience, Norwegian University of Science and Technology (NTNU), Trondheim, Norway
  2. 2Department of Neurosurgery, St Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
  3. 3Department of Physical Medicine and Rehabilitation, St Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
  4. 4Department of Diagnostic Imaging, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
  5. 5Department of Circulation and Diagnostic Imaging, Norwegian University of Science and Technology (NTNU), Trondheim, Norway
  6. 6Department of Neurosurgery, Brain and Spinal injury Center, University of California, San Francisco, USA
  1. Correspondence to Dr K G Moen, Department of Neuroscience, Norwegian University of Science and Technology (NTNU), Trondheim N-7489, Norway; kent.g.moen{at}
  1. Contributors Data collection: KGM, TS and AV. Study design: KGM, TS, GTM and AV. Data analysis: KGM, MF, VB, KAK and JR. Data interpretation: KGM, TS, GTM and AV. Writing: KGM, TS, MF,VB, KAK, JR, GTM and AV.

  • Received 16 March 2012
  • Revised 23 July 2012
  • Accepted 1 August 2012
  • Published Online First 29 August 2012


Objective To study the evolution of traumatic axonal injury (TAI) detected by structural MRI in patients with moderate and severe traumatic brain injury (TBI) during the first year and relate findings to outcome.

Methods 58 patients with TBI (Glasgow Coma Scale score 3–13) were examined with MRI at a median of 7 days, 3 months and 12 months post injury. TAI lesions were evaluated blinded and categorised into three stages based on location: hemispheres, corpus callosum and brainstem. Lesions in T2* weighted gradient echo (GRE), fluid attenuated inversion recovery (FLAIR) and diffusion weighted imaging (DWI) were counted and FLAIR lesion volumes were estimated. Inter-rater reliability score was calculated. Outcome was assessed 12 months post injury using the Glasgow Outcome Scale Extended.

Results In the initial MRI, 31% had brainstem lesions compared with 17% at 3 months (p=0.008). In the FLAIR sequences, number and volumes of lesions were reduced from early to 3 months (p<0.001). In T2*GRE sequences, the number of lesions persisted at 3 months but was reduced at 12 months (p=0.007). The number of lesions in DWI and volume of FLAIR lesions on early MRI predicted worse clinical outcome in adjusted analyses (p<0.05).

Conclusion This is the first study to demonstrate and quantify attenuation of non-haemorrhagic TAI lesions on structural MRI during the first 3 months after TBI; most importantly, the disappearance of brainstem lesions. Haemorrhagic TAI lesions attenuate first after 3 months. Only early MRI findings predicted clinical outcome after adjustment for other prognostic factors. Hence valuable clinical information may be missed if MRI is performed too late after TBI.


Traumatic axonal injury (TAI) or diffuse axonal injury is histopathologically characterised by axonal swelling and, depending on the severity, by subsequent secondary axotomy.1 In autopsy studies, TAI has been shown to be present in all cases of fatal traumatic brain injury (TBI).2 However, using more sensitive MRI techniques, TAI has been shown to be an important part of the brain injury in surviving patients with moderate and severe TBI,3 and even in mild cases.4

As TAI is grossly underestimated by CT,5 MRI has become the image modality of choice in the subacute and chronic phase of TBI.6 Recent advanced techniques, such as the diffusion tensor imaging, have consistently demonstrated loss of axonal integrity in TBI patients compared with controls,7 but normative data for diagnostic use in individual patients are not yet sufficiently developed. Therefore, conventional MRI sequences is still considered to be the standard imaging method in the routine clinical evaluation of TBI patients,8 in addition to the initial CT examinations. T2*gradient echo (T2*GRE) sequences are sensitive to iron in blood breakdown products and depict haemorrhagic TAI lesions, while fluid attenuated inversion recovery (FLAIR) also depicts non-haemorrhagic lesions due to oedema.

Early MRI findings after injury have been associated with clinical outcome.9–11 It may, however, be challenging to perform MRI during the first weeks post injury in patients who are unstable or do not cooperate. One study has demonstrated that haemorrhagic TAI lesions appear less conspicuous with time,12 while no longitudinal conventional MRI studies have assessed how non-haemorrhagic TAI lesions evolve in number and volume from the acute to the chronic stage. Such studies have been requested, and may clarify whether it is important to perform MRI early after TBI.13 ,14

Hence the aim of this study was to examine the evolution of TAI lesions during the first year post injury using conventional MRI in the early phase and at 3 and 12 months. Furthermore, we aimed to relate the findings to 12 month clinical outcome.

Material and methods


From October 2004 to July 2007, 125 individuals aged 11–65 years were admitted with moderate or severe head injury, according to the Head Injury Severity Scale,15 to the Neurosurgical Department, Trondheim University Hospital, Norway.

Of these 125 consecutive patients, 18 patients (14%) died (13 from intracranial hypertension, three from sequelae of the head injury and two from factors not associated with the injury). Nine patients (7%) refused to participate or were not able to cooperate. Patients with a premorbid condition, such as severe cerebrovascular diseases, severe psychiatric conditions or substance abuse, were excluded (n=14, 11%). Twenty-three patients (18%) could not complete all three repeated examinations because of logistic, geographical or other reasons. In three patients their early MRI scan was performed after 4 weeks (35–43 days post injury) so these patients were excluded. Thus, in total, in 58 patients (46%) all MRI scans were performed. The first scan was performed at a median of 7 days (range 0–26); the repeated MRIs were performed at 3 and 12 months after injury.


MRI was performed at the study hospital or at one of the local hospitals using a 1.5 T system (Siemens Symphony or Siemens Avanto; Siemens Medical, Erlangen Germany; the scanners are comparable). The protocol, previously described,3 consisted of sagittal TSE T2 imaging, sagittal, coronal and transverse T2 fluid attenuated inversion recovery (FLAIR) imaging, transverse T2* weighted gradient echo imaging (GRE), transverse SE T1 imaging and diffusion weighted imaging (DWI) with diffusion gradients in x, y and z dimensions and images at B=0, 500 and 1000 s/mm2 with automatically calculated apparent diffusion coefficient (ADC) maps.

To estimate total intracranial volume (TICV), we used an additional sequence not described by Skandsen et al 3: T1 MPRAGE (TR=7.1 ms; TE=3.45 ms; flip angle=7°; TI 1000 ms; FOV 256×256; acquisition matrix of 256×192×128, reconstructed to 256×256×128, giving a reconstructed voxel resolution of 1×1×1.33 mm). The sequence was missing for three patients.

Image analysis

The first author (KGM), in cooperation with two experienced neuroradiologists (KAK and JR), characterised the image findings based on visual inspection according to predefined variables (number of lesions in T2*GRE, FLAIR and DWI and volume of FLAIR lesions). All were blinded to patient identification, clinical information and time of examination.

We classified TAI lesions, using a modified staging based on the neuropathological characteristics of this type of injury, into three stages16–18: hemispheres/cerebellum (stage 1), corpus callosum (stage 2) and brainstem (stage 3). TAI lesions were counted separately in different locations and on each side of the midline in the T2*GRE, FLAIR and DWI, including the ADC map. For the analyses of lesions in DWI, all hyperintense lesions in the B-1000 image were counted. If there were >10 lesions in one of the locations unilaterally, we automatically assigned them the number 15. This was done because some patients had numerous lesions which were difficult to count. Superficial lesions in the cerebral cortex were defined as contusions.19 Isolated fornix injuries were counted as TAI stage 2 lesions, and lesions of the cerebellar peduncles were classified as brainstem injury (TAI stage 3). Parenchymal changes following insertion of drain or intracranial pressure devices were not defined as traumatic lesions. Periventricular signal hyperintensities (‘caps and bands’) were not analysed as they are considered a normal phenomenon.20 ,21 In six patients we identified white matter hyperintensities attributable to other non-traumatic comorbid non-symptomatic disease.22 These patients were excluded from the analyses involving the FLAIR sequences, but not from the other sequences as these lesions more confidently can be ascribed to the trauma.

The hypointense foci identified in T2*GRE indicating haemorrhagic lesions represent magnetic susceptibility effects due to haemoglobin degradation products.12 The size of these foci will be strongly influenced by the properties of the imaging sequence and the MR system, particularly the magnetic field strength; hence we did not evaluate volumes of these foci.

Estimation of volume of FLAIR lesions

The FLAIR sequences with visible TAI lesions were further analysed by KGM using the Brain Voyager QX software, V.1.2.23 The sagittal FLAIR sequence was loaded, and the lesions analysed based on location: hemispheres, corpus callosum, brainstem, thalamus/basal ganglia and cerebellum. Lesions in the thalamus and basal ganglia were added together because of the difficulty separating these structures in the sagittal sequence. If the sagittal sequence was missing or had poor quality, we used the coronal or transverse sequence.

The lesions were segmented manually by drawing contours around each lesion. The volume was calculated by multiplying voxel size by the number of affected voxels for each cross section. Brain Voyager automatically performed correction for gap between the cross sections.

Other authors have recommended adjustment for TICV and hence we computed TAI lesion indexes based on the TAI lesion volume divided by the TICV with a fully automated segmentation using NeuroQuant (CoreTechs Labs, California, USA).11 ,24

Inter-rater reliability

A total of 89 out of 174 MRI scans (distributed equally between the different time points) were chosen evenly and randomly from the list of cases. A third neuroradiologist (MF) evaluated these scans for the number of FLAIR, T2*GRE and DWI lesions. For the evaluation of agreement in the FLAIR volume estimation, 37 scans with non-haemorrhagic TAI lesions (19 early MRI scans and nine scans from 3 and 12 months, respectively) were reassessed. The volume analyses were performed by VB in cooperation with MF. They were blinded to the results of the other examiners as well as to patient identification, clinical information and time of examination.

Classification, injury related variables and outcome

Patients were categorised based on the Head Injury Severity Scale criteria into moderate (Glasgow Coma Scale (GCS) score of 9–13) or severe (GCS score of ≤8) head injury. GCS scores were evaluated in the emergency room or before intubation. The prospective collection of injury related variables is described in a previous study.3 CT scans of the patients were reviewed by a radiologist (IHS) and classified according to the Rotterdam CT classification, using the worst scan.25 Global outcome was assessed at 12 months post injury by telephone or personal contact with the patient and/or a close relative, using the structured interview for Glasgow Outcome Scale Extended (GOSE).26

Statistical analysis

Statistical analyses were carried out using the IBM Statistical Package for the Social Sciences (SPSS) Statistics V.19 and STATA/SE V.11.2.

We used the Student's t test for normal distributed data and the χ2 test/Fischer exact test for comparison of proportions. For skewed distributions we used the Kruskal–Wallis and Mann–Whitney U tests, while in the pairwise comparisons of related samples, we used the Wilcoxon signed rank test. Counts of lesions were fitted to a longitudinal Poisson model. We used likelihood ratio test for determination of whether a random intercept or coefficient model was most appropriate. The volume analyses were carried out using a longitudinal mixed effect model. As residuals of the volumes were not found to be normal distributed, the volumes were transformed into their third power. Residuals, random intercepts and slopes were tested against the normality distribution and by the Shapiro–Wilk W test.

The association between prognostic factors and outcome was analysed with ordinal logistic regressions, with GOSE category as the dependent variable.

An inter-rater reliability analysis was performed using the linear weighted Cohen's κ statistic to determine consistency for categorical variables. We considered κ values below 0.20 as poor, 0.21–0.40 fair, 0.41–0.60 moderate, 0.61–0.80 good and above 0.80 very good.27 For inter-rater reliability of the continuous variables, we calculated intraclass correlation coefficients (ICC) using two way random single measures (consistency/absolute agreement). An ICC >0.9 is classified as excellent, 0.4–0.9 as good and <0.4 as poor agreement.28 As ICC does not take into account whether the agreement is related to the underlying value, we also computed Bland–Altman plots. The lower and upper limits of agreement were calculated by computing the SD of the mean difference between the counts ±1.96.

The precision of the estimates was assessed with 95% CIs, and a p value <0.05 was considered statistically significant. Corrections for multiple comparisons were not performed.


Patient characteristics, occurrence and staging of TAI

Demographic, injury related variables and patient outcomes are presented in table 1.

Table 1

Patient characteristics

Fifteen patients (25%) decreased by one or two TAI stages during the 1 year follow-up. In the first MRI, 18 patients (31%) had brainstem lesions compared with 10 patients (17%, p=0.008) at 3 months (table 2).

Table 2

Evolution of staging of traumatic axonal injury over 1 year

Different MRI sequences: number of lesions and location

The number of lesions in the FLAIR sequence was significantly reduced in all locations at 3 months compared with early MRI, except for the cerebellum (table 3).

Table 3

Number of traumatic axonal injury lesions stratified into MRI sequence and location

Only four patients had visible FLAIR lesions in the cerebellum in the early MRI. Attenuation of the FLAIR lesions is also illustrated in figure 1.

Figure 1

(A–C) Sagittal midline fluid attenuated inversion recovery images of the same patient performed at three different time points (4 days, 3 months and 12 months). The first examination (A) shows lesions in the corpus callosum, thalamus and brainstem. At 3 months (B), these lesions have disappeared, but we find sequelae with incipient central atrophy. At 12 months (C), there is pronounced central and cortical atrophy. (D–F) T2*gradient echo images in the transverse plane of the same patient at the same time intervals. Multiple bilateral haemorrhagic lesions were found in the brainstem that persisted during the follow-up period of 12 months.

The highest number of lesions were depicted in the T2*GRE sequence, compared with the FLAIR (p<0.001) and the diffusion (p<0.001) sequences. For this sequence, however, there was no difference between the number of lesions in the early MRI compared with 3 months whereas the total number of T2*GRE lesions decreased significantly between 3 and 12 months post injury (table 3).

In the diffusion sequence, we observed a significantly higher number of lesions in the early MRI compared with 3 months in all locations, except for the basal ganglia and cerebellum. In the ADC map, no lesions were detected at 3 and 12 months.

In the 3 month MRI, we identified significantly fewer cortical contusions (mean 1.41 (SD 1.76)) compared with the early MRI (mean 3.26 (SD 3.17), p<0.001). No significant difference was found compared with the 12 month MRI (mean 1.29 (SD1.67), p=0.066).

FLAIR TAI volumes

The total TAI lesion volume in early MRI was significantly higher for patients with severe TBI compared with moderate TBI (p<0.001). Total lesion volumes at 3 months were significantly reduced compared with early MRI for both moderate and severe TBI. No statistically significant differences were found in total volume between 3 and 12 months. Similar results were also found for each location in both moderate and severe TBI, and most reached statistical significance (table 4).

Table 4

Total traumatic axonal injury volume in the FLAIR sequence

We also computed volumetric estimates, TAI lesion indexes, corrected for TICV. However, this did not add any new information, and total lesion volumes were used in further analyses.

Outcome analyses

Both mean number and volumes in the different outcome categories are displayed graphically in figure 2, indicating distribution of more lesions in patients with low GOSE scores. In the multiple ordinal logistic regressions with GOSE score as the dependent variable, we adjusted for the well known prognostic factors age, GCS score and pupillary dilatation (table 5).

Figure 2

Mean number of fluid attenuated inversion recovery (FLAIR), T2*gradient echo (T2*GRE) and diffusion weighted imaging (DWI) lesions and the mean FLAIR lesion volume per Glasgow Outcome Scale Extended (GOSE) category, indicating more lesions in low GOSE categories.

Table 5

Ordinal logistic regression predicting outcome category for patients with moderate and severe traumatic brain injury

Both number of T2*GRE and total FLAIR lesion volume at early and 3 month MRI predicted outcome in the univariable analyses, but after adjustment, only the number of DWI lesions and FLAIR lesion volume in the early MRI were associated with outcome.

The outcome analyses were also performed separately for moderate and severe TBI. Almost the same results were obtained regarding severe TBI, but the number of lesions depicted on T2*GRE showed only a tendency towards predicting outcome in the unadjusted analysis (p=0.057–0.059). In moderate TBI, only age (OR 1.07 (95% CI 1.02 to 1.12), p=0.003) and Rotterdam CT score (OR 3.19 (95% CI 1.38 to 7.38), p=0.007) were related to outcome in the unadjusted logistic regressions. The number of lesions in the T2*GRE (p=0.65–0.90) and FLAIR sequences (p=0.68–0.97) and FLAIR lesion volumes (p=0.52–0.53) were not associated with outcome in univariable analyses.

Inter-rater reliability

The overall inter-rater reliability of the TAI staging was good; weighted linear Cohen's κ of 0.74 (95% CI 0.69 to 0.80, p<0.001). The inter-rater agreement for number of lesions on T2*GRE sequences and volume estimation in the FLAIR sequences were excellent in the early MRI (table 6).

Table 6

Inter-rater agreement in different MRI sequences

For each MRI sequence, Bland–Altman plots showed a tendency towards less agreement for patients with numerous lesions.


In this prospective study, we have quantified to what extent conventional MRI findings of TAI attenuate over time. We have demonstrated that a significant reduction in the number and volume of non-haemorrhagic TAI lesions depicted in FLAIR sequences occurred during the first 3 months. Importantly, brainstem lesions were often no longer visible, leading to a lower stage of TAI. The number of haemorrhagic TAI lesions depicted in T2*GRE sequences attenuated after 3 months post injury.

Another important finding was that after adjustment for well known prognostic factors, only number of diffusion lesions and TAI FLAIR volume in early MRI were related to outcome. Subgroup analyses revealed that this association to GOSE score was not found in the moderate group.

Different MRI sequences: non-haemorrhagic lesions

Our study showed both the number and volume in the FLAIR sequence to be considerably reduced during the first 3 months. Previous longitudinal studies have demonstrated that TAI eventually results in cerebral atrophy,29 ,30 but no other studies have evaluated the evolution of both number and volume of FLAIR lesions longitudinally. In the acute stage, the FLAIR sequence may show non-haemorrhagic TAI lesions due to initially vasogenic and later cytotoxic oedema, the latter most predominant between 1 and 2 weeks post injury.31 Therefore, the extent of the non-haemorrhagic lesions seems to decrease as oedema gradually resolves. However, residual counts and volumes were also depicted in the chronic stages in our study, probably due to post-traumatic gliosis shown as white matter hyperintensities in the FLAIR sequence.19

We found that only 60% of patients with TAI in the brainstem (TAI stage 3) in the early MRI had brainstem lesions at 3 months post injury. This is an important finding as brainstem lesions, especially bilateral, have been strongly associated with poor outcome.9

In the diffusion sequences, our study depicted significantly fewer lesions than the T2*GRE and FLAIR sequences. In contrast, Huisman et al found DWI to be more sensitive in the visualisation of TAI lesions than FLAIR and GRE.13 However, in that study, MRI was performed within 48 h post injury while in our study half of the examinations were performed after 7 days. No previous study has evaluated longitudinally the evolution of visible lesions in DWI following TBI, but our study showed that all of the lesions visible in the ADC map had disappeared by 3 months.

A remaining question is, however, at what time point, during these first weeks, is attenuation of non-haemorrhagic lesions most prominent, and this should be a focus of future research.

Different MRI sequences: haemorrhagic lesions

We found that the haemorrhagic lesions depicted in T2*GRE attenuated after 3 months. Previous studies have claimed the T2*GRE sequence as the best method of visualising microhaemorrhagic lesions.32 ,33 It has been argued that haemorrhagic lesions may persist for years.34 The present study, however, supports the previous findings of Messori et al, who performed a longitudinal T2*GRE MRI study demonstrating that microhaemorrhagic signal foci appeared less conspicuous with time.12


The total FLAIR volumes in the early MRI were related to outcome both in the simple and adjusted logistic regressions. Earlier studies have also found that FLAIR lesion volume both in the early and later stages correlates with clinical outcome,10 ,11 but in these studies no adjustment for other prognostic factors was performed. A study by Schaefer et al, of patients examined within 48 h, also demonstrated a univariable association between FLAIR volumes and outcome, but weaker than for DWI.35

The numbers of T2*GRE lesions were associated with clinical outcome only in the univariable analyses. This is in accordance with other studies,35–37 however no adjustments for prognostic factors were conducted in these studies. In a study by Scheid et al,32 no relationship with outcome was found.

Interestingly, neither FLAIR nor T2*GRE lesions showed a tendency to predict global outcome in the subgroup analyses of moderate TBI patients. This is in line with a previous study where no association was found between depth of lesion and outcome in moderate TBI.9 This lack of association with global outcome indicates that factors other than neuroimaging findings may explain the variation in global outcome after moderate TBI. Furthermore, in many patients with moderate TBI, brain contusion, rather than TAI, could be the main lesion type and their size and location may have greater impact on outcome.9 In the supplementary data (available online only), we observed that the number of FLAIR lesions predicted information processing speed at 12 months, indicating that neuropsychological measures could be more sensitive for moderate TBI patients.

Inter-rater agreement

We experienced good inter-rater agreement in the categorisation of TAI lesions. For the counting of lesions in the T2*GRE sequences, we achieved an agreement in the excellent range, while we obtained an agreement in the high end of the ‘good’ range for the diffusion sequences. This result is contrary to the findings of Weiss et al who stated that evaluation of TAI lesions is subject to inter-rater variability.5 However, FLAIR lesions are commonly confluent,20 and the present study shows that it is difficult to distinguish between separate and multiple FLAIR lesions. We found less inter-rater agreement in counting but excellent agreement in the estimation of FLAIR volume. Thus we suggest volume estimation should be the preferred method in the assessment of FLAIR lesions.

Strengths and limitations of the study

The strengths of this study include the prospective data collection with an overview of all eligible patients. MRI was performed at three different time points during the first year in a large number of TBI patients. However, for the outcome analyses, an even larger sample size would have been preferable. We have also conducted inter-rater assessments, and all image analyses were performed blinded.

There are limitations to the study; it would have been preferable if time from injury to the first scanning was equal in all cases, and if the first MRI was performed even earlier. This, however, is difficult to obtain in a clinical cohort for both medical and logistic reasons. Detection of the small GRE lesions could be dependent of the angling of the cross section of this two-dimensional MRI sequence. In the volume calculations, the thickness of the slice and the gap between the cross sections could influence the estimated volume. Finally, the outcome assessments were not performed blind to the clinical information.


We have conducted a study of serial MRI findings, with a focus on TAI in a sample of patients with moderate and severe TBI. The most important finding was that non-haemorrhagic TAI lesions depicted in FLAIR sequences, including brainstem lesions, often disappeared during the first 3 months. In contrast, the haemorrhagic TAI lesions attenuated more slowly. Both the number of non-haemorrhagic and haemorrhagic lesions and also the volume of the non-haemorrhagic lesions were associated with global clinical outcome, but only early MRI findings were associated with outcome in multivariable analyses. This association was not found for moderate TBI patients.

The clinical implication of this study is that MRI should preferably be performed during the first few weeks post injury in severe TBI patients to improve outcome prediction. Because of the prominent early attenuation of non-haemorrhagic lesions, we recommend early MR imaging in moderate TBI also, not least in order to detect brain injury.


We thank Stine Borgen Lund and Beate Holmqvist Karlsen for help in the management of the database and the GOSE interviews, and Erik Magnus Berntsen and Ingrid Haavde Strand for their help utilising the BrainVoyager software and CT classification, respectively.


  • Funding KGM and TS have, during the study period, received a research grant from the liaison committee between the Central Norway Regional Health Authority (RHA) and the Norwegian University of Science and Technology (NTNU). VB has, during the study period, received a research grant from the National Resource Center for fMRI, Department of Diagnostic Imaging, St Olav University Hospital.

  • Competing interests None.

  • Ethics approval The local Regional Ethical Committee for Health Region Mid-Norway approved the study.

  • Provenance and peer review Not commissioned; externally peer reviewed.


Visit the full archive of podcasts for JNNP here >>

Free sample
This recent issue is free to all users to allow everyone the opportunity to see the full scope and typical content of JNNP.
View free sample issue >>

Don't forget to sign up for content alerts so you keep up to date with all the articles as they are published.