Diffusion-Weighted Imaging Improves Outcome Prediction in Pediatric Traumatic Brain Injury

Transcription

1 JOURNAL OF NEUROTRAUMA 25: (October 2008) Mary Ann Liebert, Inc. DOI: /neu Diffusion-Weighted Imaging Improves Outcome Prediction in Pediatric Traumatic Brain Injury Nicholas R. Galloway, 1 Karen A. Tong, 2 Stephen Ashwal, 3 Udochukwu Oyoyo, 2 and André Obenaus 1,2 Abstract Diffusion-weighted imaging (DWI) and consequent apparent diffusion coefficient (ADC) maps have been used for lesion detection and as a predictor of outcome in adults with traumatic brain injury (TBI), but few studies have been reported in children. We evaluated the role of DWI and ADC for outcome prediction after pediatric TBI (n 37 TBI; n 10 controls). Fifteen regions of interest (ROIs) were manually drawn on ADC maps that were grouped for analysis into peripheral gray matter, peripheral white matter, deep gray and white matter, and posterior fossa. All ROIs excluded areas that appeared abnormal on T2-weighted images (T2WI). Acute injury severity was measured using the Glasgow Coma Scale (GCS) score, and 6 12-month outcomes were assessed using the Pediatric Cerebral Performance Category Scale (PCPCS) score. Patients were categorized into five groups: (1) controls; (2) all TBI patients; (3) mild/moderate TBI with good outcomes; (4) severe TBI with good outcomes; and (5) severe TBI with poor outcomes. ADC values in the peripheral white matter were significantly reduced in children with severe TBI with poor outcomes ( mm 2 /sec) compared to those with severe TBI and good outcomes ( mm 2 /sec; p 0.05). We also found that the average total brain ADC value alone had the greatest ability to predict outcome and could correctly predict outcome in 84% of cases. Assessment of DWI and ADC values in pediatric TBI is useful in evaluating injury, particularly in brain regions that appear normal on conventional imaging. Early identification of children at high risk for poor outcome may assist in aggressive clinical management of pediatric TBI patients. Key words: apparent diffusion coefficient; children; diffuse axonal injury; magnetic resonance imaging Introduction RAUMATIC BRAIN INJURY (TBI) is a major cause of disabil- in the United States and is particularly common in Tity children and adolescents (Giza, 2006). Even mild forms of diffuse axonal injury (DAI), a common sequelae of TBI, can lead to memory loss, seizures, cognitive impairment, behavior and attention disorders, headaches, and other impairments that can prevent normal development in children (Parizel et al., 1998). While previous studies have been unable to correlate injury severity with radiological findings, diffusion-weighted imaging (DWI) has improved clinical outcome prediction in adults (Goetz et al., 2004; Hou et al., 2007; Schaefer et al., 2004). Hou et al. (2007) showed that, in adult TBI patients, there were higher apparent diffusion coefficient (ADC) values in deep gray and white matter when there was severe injury compared to those patients with favorable outcomes or controls. In other studies, detection of DAI lesions by DWI had a stronger correlation with modified Rankin scores than any other single magnetic resonance imaging (MRI) sequence for the detection of DAI lesions (Schaefer et al., 2004). Proton MR spectroscopy (MRS) and MRI volumetric analysis of pediatric TBI have also increased our understanding of the underlying mechanisms of injury and have demonstrated promise in predicting outcome but little is known about quantitative diffusion changes in the injured pediatric brain (Ashwal et al., 2006; Ichord et al., 2007; Wilde et al., 2005). DWI is sensitive to the random microscopic translation of water molecules and has proven useful in detecting acute ischemia and differentiating cytotoxic and vasogenic edema (Ashwal et al., 2006; Chien et al., 1992; Hergan et al., 2002; Kidwell et al., 1999; Suh et al., 2001). Clinically, it is very important to understand the nature of posttraumatic brain edema formation, because it is believed that some TBI lesions have MRI signal abnormalities that reflect transient, reversible edema while other abnormalities represent nonviable tissue (Takayama et al., 2000). Regardless of the type of Departments of 1 Radiation Medicine, 2 Radiology, and 3 Pediatrics, Loma Linda University, Loma Linda, California. 1153

2 1154 injury, DWI may reveal diffusion abnormalities hours or even days before computed tomography (CT) imaging can detect damaged tissues (Suh et al., 2001) and may provide a more accurate evaluation of the spatial extent of brain damage compared to conventional MRI. In a study of non-accidental pediatric neurotrauma, conventional MRI either underestimated or missed abnormalities seen with DWI in 81% of patients (Suh et al., 2001). It is possible that regions with DWI abnormality and no corresponding T2 abnormalities represent areas of salvageable tissue that, if untreated, could suffer permanent damage (Jones et al., 2000; Suh et al., 2001). DWI also allows for quantification of water mobility by means of determining ADC values, a calculated measure of the translational, random water motion as limited by cellular structures and composition. ADC measurements are particularly useful for detecting DAI, which is thought to be the primary cause of disability and poor outcome in TBI patients (Parizel et al., 1998). DAI lesions usually occur at the graywhite matter interface due to abrupt changes in tissue density and rigidity that are exacerbated when unequal rotational forces are applied during TBI (Adams et al., 1982; Hergan et al., 2002; Parizel et al., 1998). In the past, it has been difficult to detect axonal damage or ischemia in children because of the relative high water content of the immature, developing white matter of the infant and adolescent brain. The contrast between uninjured immature brain tissue and injured tissue is improved greatly with DWI, because immature white matter has less water restriction than mature tissue (hyperintense ADC signal relative to adult brain) and injured tissue typically has increased water restriction (hypointense ADC signal relative to uninjured immature brain tissue) (Suh et al., 2001). DWI is effective for detecting both hemorrhagic and nonhemorrhagic lesions. By using susceptibility-weighted imaging (SWI), we have previously shown that the presence of hemorrhagic lesions in pediatric TBI is much more common than the estimated 10 30% that T2-weighted images (T2WI) revealed in past studies (Tong et al., 2003, 2004; Yanagawa et al., 2000). The echo-planar sequences used in DWI are also much more sensitive to magnetic susceptibility than T2WI or fluid-attenuated inversion recovery (FLAIR), so DWI has the potential for improved detection of hemorrhagic lesions. While gradient recalled echo (GRE) sequences and SWI are much better in detecting hemorrhagic lesions, they are not as effective as DWI in detection of non-hemorrhagic lesions (Schaefer et al., 2004), so DWI is well suited to improve observations of both types of lesion. We evaluated ADC values in a group of children and adolescents with TBI in 15 bilateral brain regions that were then subsequently grouped into five similar brain regions. In addition, we used DWI and ADC values alone or in combination with clinical outcome measures to predict long-term neurological outcomes. Our goal was to determine whether these imaging variables were sensitive in detecting injury in the normal-appearing brain following TBI and useful for predicting 6 12-month outcomes. Methods Patient selection We studied 37 children and adolescents with TBI and presumed DAI admitted to the Intensive Care Unit (ICU) at GALLOWAY ET AL. Loma Linda University Children s Hospital between March 2001 and September 2002 (Table 1). The mean age of this cohort was years (range, years), with 25 males and 12 females. This was a cohort of patients with TBI without sufficient rapid neurological recovery in the first day that underwent MRI during acute hospitalization. We excluded children with previous head injuries, pre-existing neurological disorders, central nervous system (CNS) malformations, developmental disability, psychiatric illness, missile injuries, suspected non-accidental trauma, age younger than 1 month, and poor quality MR images. These patients were part of a larger ongoing study examining MRI for outcome prediction after various forms of acute CNS insults and were approved by our Institutional Review Board (Tong et al., 2004). Control patients (n 10) with an average age of 8.7 years (range, 1 16 years) were also included in the study. The controls consisted of five males and five females that were being imaged for other clinical indications that excluded trauma. None of the control patients had visible abnormalities on any imaging modality. A neurosurgeon or neurologist examined all patients prior to MRI acquisition. Clinical variables included age and initial Glasgow Coma Scale (GCS) score. We dichotomized the injury groups into severe (GCS 3 8) and mild/moderate (GCS 9 15) injury categories according to definitions used by the American Academy of Neurological Surgeons and the Brain Injury Foundation (Bullock et al., 1996). Outcomes were assessed using the Pediatric Cerebral Performance Category Scale (PCPCS) score, a six-point outcome scoring system modified from the Glasgow Outcome Scale (GOS) score (Fiser, 1992), which quantifies the overall functional and cognitive impairment of infants and children. This scoring system has been validated in pediatric patients after acute CNS injuries and has been shown to correlate with other measures of psychometric function (Fiser et al., 2000). The score includes the following outcomes at 6 12 months after injury: (1) normal, i.e., can perform all age-appropriate activities; (2) mild disability, i.e., conscious, alert and able to perform most age-appropriate activities but may have mild neurological deficits; (3) moderate disability, i.e., conscious, has sufficient cerebral function for age-appropriate activities of daily life, but has significant cognitive impairment; (4) severe disability, i.e., conscious, is dependent on others for daily support because of impaired brain function; (5) persistent vegetative state (PVS); and (6) death. For the purpose of analysis, the six PCPCS scores were dichotomized into two groups: (1) good outcome (PCPCS 1 2), which represents normal or mild disability and (2) poor outcome (PCPCS 3 5), which represents moderate or severe disability, or in a vegetative state. There were no patients who died after study entry. Magnetic resonance imaging When medically stable, children were transported to the MR scanner and monitored by staff from the ICU or transport team and radiology personnel. MRI was acquired at a mean of 7 4 days post-injury. Studies were performed using a circularly polarized head coil in a conventional 1.5- Tesla (T) whole body imaging system (Magnetom Vision, Siemens Medical Systems, Iselin, NJ). DWI was performed using a single-shot echo planar technique (TR/TE

3 DIFFUSION-WEIGHTED IMAGING IN PEDIATRIC TBI 1155 TABLE 1. CLINICAL DATA FOR PEDIATRIC TBI PATIENTS Patient Age (years) Sex Traumatic event Days to MRI GCS PCPCS Brain ADC a F MVA M MVA M Fell off bike M Sports related TBI M MVA F MVA M Fell off motorcycle F Fell off cliff in car F Hit by car F MVA M Assaulted F MVA M Hit by car M MVA M Hit by car M Hit by car M Hit by car F MVA-rollover M Unknown M MVA M MVA F MVA M ATV accident F MVA M MVA F MVA M Hit by car M Hit by car M MVA M MVA F Hit by car F MVA M Hit by car F Hit by car M MVA M Hit by car M MVA a Brain ADC is the average total ADC for all 30 regional values. GCS, Glasgow Coma Scale; PCPCS, Pediatric Cerebral Performance Category Scale; MVA, motor vehicle accident; ATV, all terrain vehicle; ADC, apparent diffusion coefficient ( 10 5 mm 2 /sec). 4000/110 msec, 5 mm thick) in three spatial encoding directions and two b values (500, 1000 sec/mm 2 ). Routine MRI also included axial dual spin-echo (SE; TR 2500 msec, TE 22/80 msec, 5 mm thick), sagittal T1-weighted SE (TR/TE 550/22 msec, 5 mm thick), axial FLAIR (TR/TE 9000/110 msec, 5 mm thick), and axial two-dimensional (2D) GRE (FLASH, TR/TE 500/18 msec, flip angle 15, band width 78 Hz/pixel, 4 5 mm thick) sequences. Only DWI datasets were analyzed for this study. Image analysis MR images were reviewed on a clinical workstation (DS3000, Impax; Agfa Inc.), and ADC maps were generated. Fifteen bilateral (except for the splenium and pons) regions of interest (ROIs) were manually drawn on ADC maps, using the corresponding T2WI as a reference image (Fig. 1). These regions were grouped for analysis into peripheral gray matter (PGM; regions 1, 5, 10, 13), peripheral white matter (PWM; regions 2, 4, 11, 12), deep white matter (DWM; regions 3, 6, 9), deep gray matter (DGM; regions 7, 8), and posterior fossa (PF; regions 14, 15). We also calculated the mean ADC value for all ROIs (i.e., average total ADC). ROI sizes varied by region, but were generally 1.0 cm 3 for peripheral white matter and posterior fossa, cm 3 for DWM, 0.1 cm 3 for PGM, and 0.5 cm 3 for DGM. In all cases, ROIs avoided areas with obvious lesions on corresponding T2WI. In a few individuals, ROIs could not be drawn, because T2WI showed tissue abnormalities within the entire region at all brain levels. For most scans, the plane of the T2WI and DWI slice were not the same, so a sagittal T1 image was referenced to assure consistency in the level of ROI placement. Statistical analysis Comparisons between the mean ADC values of the follow-up outcome groups and individual ROIs were evaluated

4 1156 GALLOWAY ET AL. FIG. 1. Axial apparent diffusion coefficient (ADC) maps indicating 30 regions of interest (ROI) selected for quantitative extraction of ADC values. (1) Frontal Gray Matter, (2) Frontal White Matter, (3) Corpus Callosum Body, (4) Parietal White Matter, (5) Parietal Gray Matter, (6) Corpus Callosum Genu, (7) Lentiform Nucleus, (8) Thalamus, (9) Corpus Callosum Splenium, (10) Temporal Gray Matter, (11) Temporal White Matter, (12) Occipital White Matter, (13) Occipital Gray Matter, (14) Pons, (15) Cerebellum. using a two-way analysis of variance (ANOVA). Due to the small sample sizes in the severe (n 1) and vegetative state (n 1) follow-up categories, outcome groups were dichotomized into the good and poor outcome groups for further analysis. T-tests were used to compare the differences between groups regarding mean ADC global and regional lesion measurements. All analyses were performed using the SPSS Statistics for Windows (SPSS Inc., Release 10.0, Chicago, IL). Differences were considered to be statistically significant at p Multiple linear regressions with stepwise selection were performed to determine which independent clinical variables were the best predictors of dependent ADC values. Logistic regression analysis was used to examine an association of clinical and neuroradiologic features to outcome after adjusting for gender and age. Results Clinical The mean age of the 37 TBI patients was years (range, years); control patients had a mean age of years (range, 1 16 years). Motor vehicle accidents (MVA; n 22) were the most common form of injury (Table 1); pedestrian versus MVA occurred in 12 individuals. The remaining patients were involved in a sports-related injury and an assault, and one patient fell off a bicycle. The average time to MRI after injury was 7 4 days (range, 1 to 16 days). In addition, the mean GCS score for the entire TBI patient population was 7 4 (range, 3 15), with 10 patients having a mild/moderate GCS score (i.e., 9 15) and the remaining 27 patients having a severe GCS score (i.e., 3 8). Table 2 summarizes clinical data for the following groups: (1) control patients; (2) all TBI patients; (3) patients with mild/moderate TBI (i.e., GCS 9 15) and good outcomes (PCPCS 1 2; none had poor outcomes); (4) patients with severe TBI (i.e., GCS 3 8) and good outcomes (PCPCS 1 2); and (5) patients with severe TBI (i.e., GCS 3 8) and poor outcomes (PCPCS 3 5). There were no differences in age, gender, first recorded mean arterial blood pressure, heart rate, initial ph, hematocrit or serum sodium levels between groups. The number of days in coma, days on ventilator, and days in hospital were greater in the patients with severe TBI/poor outcomes than in the mild/moderate TBI (p 0.05). The PCPCS 6 12-month outcome for the five groups were also tabulated (Table 2). Neuroimaging Comparison between different imaging modalities (T2WI, DWI) demonstrated superiority of early imaging with DWI and ADC maps. While some abnormalities could be visualized with T2WI and FLAIR in patients with poor outcome, they were not coincident with abnormalities detected on the ADC maps (Figs. 2 and 3). Similar findings were noted in patients with good outcomes and no visible T2 or FLAIR abnormalities. These examples show that DWI was able to reveal a greater extent and degree of abnormality than T2 or FLAIR. ADC values for each patient were extracted from 2 ROIs to estimate the ability for prediction of final neurological outcomes. In patients with severe injury, mean ADC values in four brain regions (frontal white matter, temporal gray matter, temporal white matter, and basal ganglia) were significantly lower in poor compared to good outcome patients (Table 3). Comparison of hemispheric differences within TBI patients revealed only four brain regions (frontal cortex, parietal cortex, occipital white matter, cerebellum) that exhibited significant differences between right and left hemispheres (data not shown). Brain regions were categorized into six subgroups: peripheral gray, peripheral white matter, deep gray and white matter, posterior fossa, and average total. They were compared in five groups: controls, all TBI, mild/moderate TBI with good outcome, severe TBI with good outcome, and se-

5 DIFFUSION-WEIGHTED IMAGING IN PEDIATRIC TBI 1157 TABLE 2. CLINICAL VARIABLES OF DICHOTOMIZED TBI PATIENTS Controls All TBI Mild/moderate TBI with Severe TBI with good Severe TBI with poor (n 10) (n 37) good outcome a (n 10) outcome (n 19) outcome (n 8) Age (years) Gender (M/F) 5/5 25/12 6/4 14/5 5/3 GCS NA ,2, ,3 MAP (mm Hg) NA HR (BPM) NA ph NA Glucose (mg/dl) NA Hematocrit NA Serum sodium (meq/l) NA Days in coma NA Days on ventilator NA Days in hospital NA Outcomes (PCPCS) Normal Mild Moderate Severe Vegetative state a Outcome defined by PCPCS. Data presented as mean SD. GCS Dichotomized: mild/moderate (9 15), severe (3 8). Significance (p 0.05) for the following groups: 1 mild TBI versus controls; 2 mild TBI versus severe TBI with good outcome; 3 mild TBI versus severe TBI with poor outcome. vere TBI with poor outcome (Table 4). The average ADC values were significantly lower in deep white matter when comparing all TBI patients to controls. We were most interested in determining whether regional ADC values could discriminate between the good and poor PCPCS outcomes in the severe TBI patients. ADC differences were present only in the peripheral white matter between severe TBI and good outcome ( ) and severe TBI with poor outcome ( ; p 0.05). We also found significant differences between control patients and the severe TBI with poor outcomes group in the peripheral white matter and deep white matter. It should be noted that ADC values in the deep white matter were consistently much lower than in other brain regions (Table 4, Fig. 4). However, due FIG. 2. Comparative images between axial T2-weighted, fluid-attenuated inversion recovery (FLAIR) and apparent diffusion coefficient (ADC) maps in two patients with different outcomes. In both of these patients, the Glasgow Coma Scale (GCS) score was 3. However, the patient with poor outcome revealed visible lesions on T2-weighted imaging (T2WI) and FLAIR images, but these lesions were not coincident with the alterations observed on the ADC map (arrows) showing visible diffuse axonal injury (DAI) lesions. In the patient with a good outcome, no visible lesions were observed on the axial T2WI nor FLAIR images, but a small lesion was visible on the axial ADC map image (arrow).

6 1158 GALLOWAY ET AL. FIG. 3. Utility of diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) maps for improved detection of lesions in pediatric patients. (A) Imaging from a 17-year-old patient whose follow-up outcome was normal. He had a Glasgow Coma Scale (GCS) score of 3. Both the T2 scan and ADC map revealed a lesion in the splenium, but the T2 signal is hyperintense (suggesting edema) while the ADC signal is hypointense (suggesting restricted water diffusion). (B) An example of the improved ability of DWI to visualize abnormalities after traumatic brain injury (TBI). Images are from a 9-year-old boy who had a GCS of 3, but had moderate disability at follow-up. White matter in the T2-weighted imaging (T2WI) appears essentially normal in the contralateral hemisphere (arrows) and in the frontal lobe of the ipsilateral hemisphere (arrow) as well, but the ADC map shows extensive hypointense signal, indicating restricted water diffusion and cellular damage. TABLE 3. REGIONAL BRAIN MEAN ADC VALUES Mild/moderate with Severe TBI with Severe TBI with Brain region Controls All TBI good outcome good outcome poor outcome Frontal gray matter Frontal white matter ,2 Parietal gray matter Parietal white matter Temporal gray matter Temporal white matter Occipital gray matter Occipital white matter Basal ganglia Thalamus Cerebellum Pons (midline) CC-body CC-genu CC-splenium (midline) Mean ADC SD expressed as 10 5 mm 2 /sec. Significance (p 0.05) determined using a one-way ANOVA with a Scheffe post hoc test: 1 severe TBI and good outcome versus severe TBI with poor outcome; 2 controls versus severe TBI with poor outcome; 3 mild TBI versus severe TBI with poor outcome. CC, corpus callosum.

7 DIFFUSION-WEIGHTED IMAGING IN PEDIATRIC TBI 1159 TABLE 4. COMPARISON OF BRAIN ADC VALUES WITHIN FIVE PRIMARY BRAIN REGIONS Mild/moderate with Severe TBI with Severe TBI with Brain region Controls All TBI good outcome good outcome poor outcome Peripheral gray matter Peripheral white matter ,2 Deep gray matter Deep white matter , Posterior fossa Average total brain Mean ADC SD expressed as 10 5 mm 2 /sec. GCS Dichotomized: mild/moderate (9 15), severe (3 8). ADC values were significantly different (p 0.05) using the one-way ANOVA with a Scheffe post hoc test: 1 severe TBI and good outcome versus severe TBI with poor outcome; 2 controls versus severe TBI with poor outcome; 3 mild TBI versus severe TBI with poor outcome; 4 controls versus all TBI. to the relatively small number of patients, comparisons between the mild and two severe groups did not show significance. The mean ADC value in deep white matter ( ) was also lower than the peripheral white matter in the severe TBI with poor outcomes ( ; Fig. 4). Finally, and surprisingly, patients with mild/moderate TBI and good outcomes had lower peripheral gray matter ADC values ( ) than the severe TBI and good outcome group ( ). Neuroimaging and clinical correlations Further analysis using average total brain ADC values (Table 4) and clinical variables (Table 2) revealed a significant correlation between number of days in hospital and ADC (r 0.37, p 0.02; Pearson correlations). Linear and logistic regression analyses were performed for additional comparisons of ADC and clinical values (Table 5). The r 2 value represents the proportion of variance in the dependent variables (i.e., average total ADC) that can be predicted from the different combinations of independent clinical variables. The only clinical combination that reached significance (p 0.028) was the combination of GCS score, days in hospital and heart rate (r ). This value indicates that 24.5% of the variance in the average total ADC can be predicted from these combinations of variables. The logistic regression analysis examined the ability of individual or different combinations of clinical and neuroimaging values to best predict outcome (left data column). The middle data column lists the increase in the percentage cases predicted correctly compared to the null model which represents the predictive ability of all other variables pooled together. Four individual or groups of variables reached significance: (1) GCS alone; (2) days in coma, in hospital, or on ventilator; (3) average total brain ADC; and (4) ph, glucose, days in hospital, parietal white matter ADC, and average total ADC. The average total brain ADC value (83.8%) had the largest increase (5.4%) above the pooled data to predict outcome. Discussion FIG. 4. In the pediatric brain, apparent diffusion coefficient (ADC) maps can provide excellent differential outcome after traumatic brain injury (TBI). Comparison of deep gray and deep white matter regions between controls, mild/moderate TBI with good outcomes, and severe TBI with good and poor outcomes (PCPCS) demonstrated that deep white matter was more sensitive to long-term tissue alterations. A significant decrease in ADC was observed between controls and severe TBI patients with poor outcomes (*p 0.05) that was not seen in good outcome TBI patients. The current study examined the relation between regional and total brain ADC values in reflecting the severity of injury as determined by (1) the GCS score at admission, and (2) late (6 12 month) outcomes (PCPCS scores) in children with TBI. Our most important observation was that ADC values in the peripheral white matter were significantly reduced in children with severe TBI and poor outcome ( ) compared to those with severe TBI and good outcome ( ; p 0.05). We also found that the average total brain ADC values alone had the greatest ability to predict outcome and could correctly predict outcome in 83.8% of cases. Such data would be helpful in independently and objectively evaluating a child s prognosis for recovery when comparing clinical variables such as the GCS score, which in the current study correctly predicted outcome in 81% of the TBI cases.

8 1160 GALLOWAY ET AL. TABLE 5. REGRESSION ANALYSIS OF TOTAL BRAIN ADC TO CLINICAL VARIABLES Total brain ADC: linear regression analysis Variables r 2 F-value p-value Age, GCS Sodium, heart rate, mean arterial pressure, ph hematocrit, glucose Days in coma, days on ventilator, days in hospital GCS, days in hospital, heart rate * Outcome: logistic regression analysis Variables % cases a % Change in predicted b p-value Age GCS * Age and GCS MAP, HR, ph, glucose, hematocrit, serum sodium Days in coma, in hospital, and on ventilator * PWM, PGM, DWM, DGM, posterior fossa Total brain ADC * PWM ADC * PGM ADC DWM ADC Posterior fossa ADC DGM ADC ph, glucose, days in hospital, PWM ADC, total brain ADC a Percentage of cases predicted. b Change in the percentage of cases predicted correctly above that expected using all other variables. R 2 is the proportion of variance in the dependent variable (ADC or PCPCS outcome) that can be predicted from the independent variables. The F-value is the mean square regression model divided by the mean square residual, yielding f The p-value associated with this F- value is given in column 3. *Significance is at p MAP, mean arterial blood pressure; HR, heart rate; PWM, parietal white matter; PGM, parietal gray matter; DWM, deep white matter; DGM, deep gray matter. Although peripheral white matter showed the highest correlation with outcomes, ADC values in the deep white matter were consistently lower in all TBI groups. These findings indicate greater restriction of water movement within the deep white matter in our TBI patients. This finding is consistent with previous observations that deep white matter injury is common and more severe after pediatric TBI (Adams et al., 1982; Tong et al., 2004). One result that is difficult to explain is that patients with mild/moderate TBI and good outcomes had lower peripheral gray matter ADC values ( ) than the severe TBI with good outcome group ( ). It is believed that cerebral edema occurs more commonly after pediatric TBI (Aldrich et al., 1992; Marmarou et al., 2006). Our findings would suggest that there may be differences between mechanisms responsible for edema formation between gray and white matter. Edema formation in gray matter may not be as ominous a predictor of outcome as the presence of edema in white matter. While edema formation in white matter is more likely related to the severity of diffuse axonal injury which is also more commonly associated with long-term sequelae; edema formation in gray matter may be related to transient changes in water movement that possibly could be related to glutamate release, brief periods of hypotension or acute post-traumatic seizures (Castejon et al., 2002; Kuroiwa et al., 1994; Unterberg et al., 2004). An additional potential mechanism could be ongoing cellular remodeling that is only evident in the severely injured patients with good outcomes but complete or non-existent in the severe TBI group who had poor outcomes. Thus, the increased ADC in the severe with good outcomes group could be a reflection of brain repair mechanisms, but this suggestion awaits further research. Reports of ADC changes in adult TBI patients have been divergent. In one study of 44 patients, ADC values in normal appearing brain were similar to controls, while decreased ADC values were observed in patients with brain swelling but with normal cerebral blood flow (Marmarou et al., 2006). Other investigators have found elevated ADC in normal appearing white matter in an adult TBI group with increasing ADC values with worsening injury severity (Goetz et al., 2004) rather than the decrease in ADC that we observed. It is possible that use of algorithm-based ROI selection in the study by Goetz et al. (2004) as opposed to our user-selected ROIs could account for some of these differences. However, it is more likely that the differences are due to the number of patients that were involved in MVA in their study (25%) compared to our study (60%). MVA is believed to be the major cause of DAI (Tong et al., 2004). The variability of ADC measurements is highlighted in a study of 100 adult patients with TBI and DAI, where decreased ADC

9 DIFFUSION-WEIGHTED IMAGING IN PEDIATRIC TBI 1161 was reported in 64% of lesions, increased in 34%, and normal ADC values in 12% of lesions (Schaefer et al., 2004). These findings presumably are related to the severity of injury and the timing of MRI acquisition. Our recent ADC study in a group of adults with TBI showed similar findings, in that ADC values in two brain regions (deep gray and deep white matter) had significantly lower values in the poor outcome (deep gray ; deep white ) compared to the good outcome group (deep gray , p ; deep white , p 0.031) (Hou et al., 2007). In this study, we did not determine an average total brain ADC value for comparison. However, the brain regions that showed these ADC differences in adults (deep gray matter, deep white matter) were different than the peripheral white matter involvement observed in children in the current study, again suggesting that there are likely maturational, biomechanical, physiological, and metabolic developmental differences in the response to injury. A major contributing factor likely relates to the regional differences in white matter maturation and structure in children. In previous studies, anisotropy measurements in compact white matter structures (corpus callosum, internal capsule, cerebral peduncle) were higher than those in noncompact white matter (corona radiata and peripheral white matter) (McGraw et al., 2002). Likewise, although myelination is initially greater in compact white matter, the change in myelination appeared larger in non-compact white matter during the first few years after infancy. Myelination leads to progressive tissue compartmentalization, which is another factor that restricts the mobility of water molecules (Prayer et al., 2003). Thus, one would expect that, as the deeper brain structures myelinate first, one might see lower ADC values at baseline and in response to injury in these regions, similar to those observed in our study Our previous findings, that more DAI lesions are hemorrhagic than once thought, presents potential limitations about the reliability of ADC values measured in previous studies. Blood products are known to create magnetic susceptibility effects that alter ADC values. Based on results of our susceptibility weighted imaging studies in which the number of hemorrhagic DAI lesions was approximately six times greater than that seen with conventional imaging, it is likely that many areas that had obvious lesions on ADC maps were hemorrhagic, and thus likely to lower previously reported ADC values (Tong et al., 2003). However, many areas of observed hypointensity in white matter tracts on ADC maps were so extensive and diffuse that significant lowering of ADC values due to blood products may be less likely. In a future study, we hope to correlate our SWI findings with DWI findings to further understand this relation. The current study has a number of limitations which do not materially change the predictive nature of our findings. The study could have been strengthened by use of prospective data sets from pediatric TBI patients that would have potentially yielded an increase in the predictive power of the study, particularly for the good outcome group compared to controls. In addition, the relatively high b value (1000 sec/mm 2 ) may decrease the sensitivity to diffusion changes in the pediatric brain. Similar concerns have been reported in the pediatric stroke literature (Jones et al., 2003). A lower b value (i.e., sec/mm 2 ) may also have improved sensitivity. A final potential limitation is that some ROIs may have included portions of the gray white matter junction due to partial volume effects and potentially alter the true ADC. It is important to note that diffusion tensor imaging (DTI) is now being investigated for neurotrauma assessment, particularly in white matter injury. Recent studies have reported decreases (Okanishi et al., 2007; Wilde et al., 2006; Wozniak et al., 2007) and increases (Bazarian et al., 2007) in fractional anisotropy (FA), a measure of water directionality. Ongoing research into DTI will determine its usefulness for outcome prediction in pediatric TBI. In conclusion, the present study demonstrates that the DWI and ADC signatures obtained at the time of injury, particularly in white matter, are predictive of long-term outcome in pediatric patients. Additional studies correlating these with other imaging parameters such as spectroscopy, SWI, and DTI will be helpful in objectively quantifying injury severity and will be useful to assess evolution of injury and repair and the effectiveness of treatment. Acknowledgments We thank Anna Smith for assistance with manuscript preparation. Author Disclosure Statement This study was funded in part by the UC Neurotrauma Program. References Adams, J.H., Graham, D.I., Murray, L.S., and Scott, G. (1982). Diffuse axonal injury due to nonmissile head injury in humans: an analysis of 45 cases. Ann. Neurol. 12, Aldrich, E.F., Eisenberg, H.M., Saydjari, C., et al. (1992). Diffuse brain swelling in severely head-injured children. A report from the NIH Traumatic Coma Data Bank. J. Neurosurg. 76, Ashwal, S., Holshouser, B.A., and Tong, K.A. (2006). Use of advanced neuroimaging techniques in the evaluation of pediatric traumatic brain injury. Dev. Neurosci. 28, Bazarian, J.J., Zhong, J., Blyth, B., Zhu, T., Kavcic, V., and Peterson, D. (2007). Diffusion tensor imaging detects clinically important axonal damage after mild traumatic brain injury: a pilot study. J. Neurotrauma 24, Bullock, R., Chesnut, R.M., Clifton, G., et al. (1996). Guidelines for the management of severe head injury. Brain Trauma Foundation. Eur. J. Emerg. Med. 3, Castejon, O.J., Castejon, H.V., Zavala, M., Sanchez, M.E., and Diaz, M. (2002). A light and electron microscopic study of oedematous human cerebral cortex in two patients with posttraumatic seizures. Brain Inj. 16, Chien, D., Kwong, K.K., Gress, D.R., Buonanno, F.S., Buxton, R.B., and Rosen, B.R. (1992). MR diffusion imaging of cerebral infarction in humans. AJNR Am. J. Neuroradiol. 13, Fiser, D.H. (1992). Assessing the outcome of pediatric intensive care. J. Pediatr. 121, Fiser, D.H., Long, N., Roberson, P.K., Hefley, G., Zolten, K., and Brodie-Fowler, M. (2000). Relationship of pediatric overall performance category and pediatric cerebral performance category scores at pediatric intensive care unit discharge with outcome measures collected at hospital discharge and 1- and 6-month follow-up assessments. Crit. Care Med. 28,

10 1162 Giza, C. (2006). Traumatic Brain Injury in Children, in: Pediatric Neurology: Principles & Practice, 4 th ed. K.F. Swaiman, S. Ashwal, and D.M. Ferriero (eds), Elsevier Mosby: St. Louis, MO, pps Goetz, P., Blamire, A., Rajagopalan, B., Cadoux-Hudson, T., Young, D., and Styles, P. (2004). Increase in apparent diffusion coefficient in normal appearing white matter following human traumatic brain injury correlates with injury severity. J. Neurotrauma 21, Hergan, K., Schaefer, P.W., Sorensen, A.G., Gonzalez, R.G., and Huisman, T.A. (2002). Diffusion-weighted MRI in diffuse axonal injury of the brain. Eur. Radiol. 12, Hou, D.J., Tong, K.A., Ashwal, S., et al. (2007). Diffusionweighted magnetic resonance imaging improves outcome prediction in adult traumatic brain injury. J. Neurotrauma 24, Ichord, R.N., Naim, M., Pollock, A.N., Nance, M.L., Margulies, S.S., and Christian, C.W. (2007). Hypoxic-ischemic injury complicates inflicted and accidental traumatic brain injury in young children: the role of diffusion-weighted imaging. J. Neurotrauma 24, Jones, R.A., Palasis, S., and Grattan-Smith, J.D. (2003). The evolution of the apparent diffusion coefficient in the pediatric brain at low and high diffusion weightings. J. Magn. Reson. Imaging 18, Jones, D.K., Dardis, R., Ervine, M., et al. (2000). Cluster analysis of diffusion tensor magnetic resonance images in human head injury. Neurosurgery 47, Kidwell, C.S., Alger, J.R., Di, S.F., et al. (1999). Diffusion MRI in patients with transient ischemic attacks. Stroke 30, Kuroiwa, T., Ueki, M., Chen, Q., Suemasu, H., Taniguchi, I., and Okeda, R. (1994). Biomechanical characteristics of brain edema: the difference between vasogenic-type and cytotoxictype edema. Acta Neurochir. Suppl. (Wien) 60, Marmarou, A., Signoretti, S., Fatouros, P.P., Portella, G., Aygok, G.A., and Bullock, M.R. (2006). Predominance of cellular edema in traumatic brain swelling in patients with severe head injuries. J. Neurosurg. 104, McGraw, P., Liang, L., and Provenzale, J.M. (2002). Evaluation of normal age-related changes in anisotropy during infancy and childhood as shown by diffusion tensor imaging. AJR Am. J. Roentgenol. 179, Okanishi, T., Saito, Y., Fujii, S., et al. (2007). Low signal intensity and increased anisotropy on magnetic resonance imaging in the white matter lesion after head trauma: unrecognized findings of diffuse axonal injury. J. Neurol. Sci. 263, Parizel, P.M., Ozsarlak, O., Van Goethem, J.W., et al. (1998). Imaging findings in diffuse axonal injury after closed head trauma. Eur. Radiol. 8, GALLOWAY ET AL. Prayer, D., and Prayer, L. (2003). Diffusion-weighted magnetic resonance imaging of cerebral white matter development. Eur. J Radiol. 45, Schaefer, P.W., Huisman, T.A., Sorensen, A.G., Gonzalez, R.G., and Schwamm, L.H. (2004). Diffusion-weighted MR imaging in closed head injury: high correlation with initial glasgow coma scale score and score on modified Rankin scale at discharge. Radiology 233, Suh, D.Y., Davis, P.C., Hopkins, K.L., Fajman, N.N., and Mapstone, T.B. (2001). Nonaccidental pediatric head injury: diffusion-weighted imaging findings. Neurosurgery 49, Takayama, H., Kobayashi, M., Sugishita, M., and Mihara, B. (2000). Diffusion-weighted imaging demonstrates transient cytotoxic edema involving the corpus callosum in a patient with diffuse brain injury. Clin. Neurol. Neurosurg. 102, Tong, K.A., Ashwal, S., Holshouser, B.A., et al. (2004). Diffuse axonal injury in children: clinical correlation with hemorrhagic lesions. Ann. Neurol. 56, Tong, K.A., Ashwal, S., Holshouser, B.A., et al. (2003). Hemorrhagic shearing lesions in children and adolescents with posttraumatic diffuse axonal injury: improved detection and initial results. Radiology 227, Unterberg, A.W., Stover, J., Kress, B., and Kiening, K.L. (2004). Edema and brain trauma. Neuroscience 129, Wilde, E.A., Chu, Z., Bigler, E.D., et al. (2006). Diffusion tensor imaging in the corpus callosum in children after moderate to severe traumatic brain injury. J. Neurotrauma 23, Wilde, E.A., Hunter, J.V., Newsome, M.R., et al. (2005). Frontal and temporal morphometric findings on MRI in children after moderate to severe traumatic brain injury. J. Neurotrauma 22, Wozniak, J.R., Krach, L., Ward, E., et al. (2007). Neurocognitive and neuroimaging correlates of pediatric traumatic brain injury: a diffusion tensor imaging (DTI) study. Arch. Clin. Neuropsychol. 22, Yanagawa, Y., Tsushima, Y., Tokumaru, A., et al. (2000). A quantitative analysis of head injury using T2*-weighted gradientecho imaging. J. Trauma 49, Address reprint requests to: Andre Obenaus, Ph.D. Radiobiology Program, CSP A1010 Department of Radiation Medicine Loma Linda University Loma Linda, CA aobenaus@dominion.llumc.edu