Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Prediction of brain age suggests accelerated atrophy after traumatic brain injury

Prediction of brain age suggests accelerated atrophy after traumatic brain injury RESEARCH ARTICLE Prediction of Brain Age Suggests Accelerated Atrophy after Traumatic Brain Injury James H. Cole, PhD, Robert Leech, PhD, and David J. Sharp, PhD, for the Alzheimer’s Disease Neuroimaging Initiative Objective: The long-term effects of traumatic brain injury (TBI) can resemble observed in normal ageing, suggesting that TBI may accelerate the ageing process. We investigate this using a neuroimaging model that predicts brain age in healthy individuals and then apply it to TBI patients. We define individuals’ differences in chronological and predicted structural "brain age," and test whether TBI produces progressive atrophy and how this relates to cognitive function. Methods: A predictive model of normal ageing was defined using machine learning in 1,537 healthy individuals, based on magnetic resonance imaging–derived estimates of gray matter (GM) and white matter (WM). This ageing model was then applied to test 99 TBI patients and 113 healthy controls to estimate brain age. Results: The initial model accurately predicted age in healthy individuals (r5 0.92). TBI brains were estimated to be "older," with a mean predicted age difference (PAD) between chronological and estimated brain age of 4.66 years (610.8) for GM and 5.97 years (611.22) for WM. This PAD predicted cognitive impairment and correlated strongly with the time since TBI, indicating that brain tissue loss increases throughout the chronic postinjury phase. Interpretation: TBI patients’ brains were estimated to be older than their chronological age. This discrepancy increases with time since injury, suggesting that TBI accelerates the rate of brain atrophy. This may be an important factor in the increased susceptibility in TBI patients for dementia and other age-associated conditions, motivating fur- ther research into the age-like effects of brain injury and other neurological diseases. ANN NEUROL 2015;77:571–581 7,8 raumatic brain injury (TBI) causes long-term struc- cognitive impairment, and brain volume loss. Insults, Ttural and functional alterations to the brain. Some of such as TBI, may trigger a sequence of neurobiological 1,2 these changes are thought to be progressive in nature, events that alter that trajectory, prematurely causing brain and potentially underlie the increased risk for early cogni- atrophy, and potentially manifesting as an early onset of 3 4 9 tive decline and dementia observed in TBI patients. Sim- neurodegeneration. As illustrated in Figure 1, an envi- ilar behavioral and anatomical changes are also associated ronmental insult like TBI might cause a one-off increase 5,6 with normal ageing, raising the possibility that the in apparent "brain age," or could result in an ongoing chronic consequences of TBI may contribute to the prema- interaction between injury and ageing-related or other ture development of age-associated changes to the brain. neurodegenerative processes that cause progressive brain Normal ageing can be considered as the progression atrophy. In the latter case, as more time passes since the along a temporal trajectory, where individuals gradually TBI occurred, the greater the discrepancy between chro- accumulate pathologies associated with physical decline, nological age and estimated brain age will be. This A complete listing of Alzheimer’s Disease Neuroimaging Initiative investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf View this article online at wileyonlinelibrary.com. DOI: 10.1002/ana.24367 Received Aug 21, 2014, and in revised form Jan 6, 2015. Accepted for publication Jan 19, 2015. Address correspondence to Dr Cole, Burlington Danes Building, Hammersmith Hospital, Imperial College London, London, W12 0NN, United Kingdom. E-mail: [email protected] From the and Computational, Clinical, and Cognitive Neuroimaging Laboratory, Department of Medicine, Imperial College London, London, United Kingdom Additional Supporting Information may be found in the online version of this article. V 2015 The Authors Annals of Neurology published by Wiley Periodicals, Inc. on behalf of American Neurological Association. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. 571 ANNALS of Neurology teristic changes to cognition seen during normal ageing affect the domains of executive function, memory, and information processing speed. TBI patients show a simi- lar pattern of cognitive deficits, further suggesting links between TBI and ageing of the brain. Here we developed and tested a predictive model of brain age. Machine learning techniques were used to define a model that accurately predicted chronological age in healthy individuals. The model was then applied to brain images from TBI patients, allowing a prediction of their brains’ ages to be made. It was expected that TBI patients’ brains would be older than their chronological age and that this discrep- ancy would increase with longer time since injury (TSI), FIGURE 1: Model of premature brain ageing in traumatic brain reflecting a progressive atrophy of brain tissue. Furthermore, injury. Illustration of the conceptual framework for the investi- gation of brain age in traumatic brain injury (TBI). The short- we hypothesized that the discrepancy between chronological dashed line represents the trajectory of healthy ageing as age age and predicted brain age would be reflected in cognitive (x-axis) increases, against a background gradient of increasing changes that would be consistent with age-related cognitive susceptibility to age-related pathology (y-axis), such as cogni- tive decline and dementia. Occurrence of TBI is indicated (black impairment normally seen in older individuals. arrow), with acute pathology causing an immediate departure from a healthy brain state. Two alternative brain ageing trajec- Subjects and Methods tories post-TBI are shown. The long-dashed "additive effects" line depicts a trajectory assuming a one-off hit, with damage Participants leading to the patient’s brain structure resembling an older individual, followed by a normal rate of subsequent ageing. TRAINING SET. T1-weighted magnetic resonance imaging The dash–dot "interactive effects" line represents an acceler- (MRI) data from 1,537 healthy controls were obtained from 8 ated rate of brain atrophy caused by TBI and an interaction publically accessible neuroimaging initiatives (Supplementary with normal ageing processes, with the discrepancy between Table). This provided an unbiased source of data with which to normal ageing and pathological changes increasing the greater the time since injury (TSI). Comparing predicted age difference train the age prediction model that was entirely independent (PAD) scores (i; dashed black line) and (ii; solid black line)illus- from the TBI and control test data sets. Controls had no his- trates how a greater PAD score would be expected under the tory of significant neurological or psychiatric problems, with interactive effects model with accelerating atrophy (i), com- further specific recruitment criteria made by each independent pared to the added effects model (ii), at equivalent TSI (figure study. Exclusions were made due to poor data quality leading adapted from Smith and colleagues ). [Color figure can be viewed in the online issue, which is available at wileyonline to image processing failure, identified during imaging quality library.com.] assessment. All training set data had been previously anony- mized, and ethical approval and informed consent were obtained by each specific study. possibility is consistent with the progressive decline asso- ciated with TBI, even years after injury, as demonstrated TEST SET. Ninety-nine patients with persistent neurological 10 11–14 1 by neuropsychological, neuroimaging, and animal problems after TBI (72 males, mean age6 standard deviation research. [SD]: 37.986 12.43 years) were recruited (Table 1). A comparison group of 113 healthy controls assessed on the same scanner was Using neuroimaging, it is possible to predict age in used to validate the accuracy of the age prediction model (49 healthy individuals, allowing the discrepancy between males, 43.36 20.24 years). All patients were scanned at least 1 chronological age and predicted brain age to be calculated. month post-injury (mean5 28.4 months), with a range of 1 to In clinical samples, this discrepancy can be considered as 563 months. This variation allowed us to examine the influence of an index of the deviation from a normal ageing trajectory. TSI on brain changes. The severity of TBI was classified using the Estimates of brain age, derived using machine learning, Mayo Classification System criteria, with 17% being classified as have previously been used in a number of contexts. These mild (probable) and 83% being moderate/severe. Cause of injury include measuring normal brain maturation during devel- was reported as follows: road–traffic accident (RTA; 39%), fall 16,17 opment, predicting conversion from mild cognitive (24%), assault (23%), sport-related injury (6%), other (7%). Cere- impairment to Alzheimer disease, and in neurodevelop- bral contusions were identified in 52 patients (53%) by an experi- mental disorders such as schizophrenia and borderline per- enced neuroradiologist. Exclusion criteria were as follows: sonality disorder, where patients were shown to have psychiatric or neurological illness, previous traumatic brain injury, antiepileptic medication, current or previous drug or alcohol abuse, apparently "older" brains. The deviation from normal MRI contraindication. All participants gave written informed con- ageing may reflect important neurological changes relating sent, and the local ethics committee approved the study. to clinical features such as cognitive impairment. Charac- 572 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI TABLE 1. Details of TBI Patients Characteristic TBI Patients Controls No. 99 113 Age, yr 37.986 12.43 43.36 20.24 Sex, M/F 72/27 49/64 Time since TBI, mo 28.46 63.57 — Mechanism of injury, RTA/fall/assault/sports/other 39/24/23/6/7 — Mayo Clinic criteria, probable–mild/moderate–severe 17/82 — Patient has contusions on MRI, present/absent 52/47 — GCS 9.716 4.75 — Patient medicated at first visit, medicated/not medicated 55/44 — Values are reported in mean6 standard deviation or absolute numbers. F5 female; GCS5 Glasgow Coma Scale; M5 male; MRI5 magnetic resonance imaging; RTA5 road–traffic accident; TBI5 traumatic brain injury. MACHINE LEARNING PREDICTION OF AGE. Machine Neuropsychological testing was carried out using a stand- learning analysis (see Fig 2B) was conducted using the Pattern ardized battery of tests previously shown to be sensitive to cog- Recognition for Neuroimaging Toolbox (Pronto ) and run on nitive abnormalities in TBI. These were the Trail Making GM and WM separately. Data from all subjects were converted Task, Stroop color naming and word reading, Wechsler Abbre- to a similarity matrix kernel to improve computational effi- viated Scale for Intelligence (WASI) matrix reasoning and simi- ciency in Pronto by generating a vector representation of voxel- larities subscales, Choice Reaction task, People and Doors wise intensity levels for all data and calculating the dot immediate recall test, letter fluency, and task inhibition/ product between each image. Next, the data were mean- switching. centered and a Gaussian Processes Regression (GPR) model was Procedures defined using age as the dependent variable and the similarity A high-level overview of the methods is provided in Figure 2. matrix of imaging data as the independent variables. GPR is a Methods are summarized below. machine learning extension of the classical regression model, which incorporates nonlinear and Gaussian probabilistic ele- MRI ACQUISITION PARAMETERS. For the training set of ments to allow quantitative predictions to be made using con- high-resolution T1-weighted magnetic resonance images, multi- tinuous variables. ple scanners were used, including different vendors, field strengths, and acquisition protocols (see Supplementary Table). MODEL VALIDATION. Model validation (see Fig 2C) pro- For the test set, T1 images were acquired using a Philips 3T ceeded in 3 stages to ensure independence between training and Intera scanner (Philips Medical Systems, Best, the Netherlands) test sets and to enable an unbiased demonstration of model gener- with the following parameters: matrix size5 208 3 208, slice alizability. The first stage involved running 10-fold cross-validation thickness5 1.2mm, 0.94 x 0.94mm in-plane resolution, 150 on the training set, to determine the accuracy of the model. This slices, repetition time5 9.6 milliseconds, echo time5 4.5 milli- entailed randomly selecting one-tenth of the training data to be a seconds, flip angle5 8 . temporary test set, with the remainder used for model definition. Using the learned pattern from this training set, age was predicted IMAGE PREPROCESSING. All MRI data were preprocessed on the temporary test set. This process was iterated until all images (see Fig 2A) using the Statistical Parametric Mapping (SPM8) had been included in the test set and had an age value predicted. software package (University College London, London, UK). This was followed by permutation testing with 1,000 randomiza- This included tissue segmentation into gray matter (GM) and tions to derive a p-value for each model. The second stage involved white matter (WM) maps, then registration using the nonlinear validating the model on the independent controls test data set. DARTEL algorithm to Montreal Neurological Institute space Here, the model was defined using the entire training set and resampling with a 4mm smoothing kernel. Each tissue class (N5 1,537) and then used to predict the brain ages for the control (ie, GM and WM) was processed independently after segmenta- test data set. Finally, the trained model was applied to estimate pre- tion. The preprocessing procedure ensured that all images were dicted brain ages for the TBI patients (see Fig 2D). well aligned and appropriate for voxelwise analysis at the machine learning stage. April 2015 573 ANNALS of Neurology FIGURE 2: Overview of the study methods. Study data comprised 2 sets, training and test. The training set used structural magnetic resonance imaging from 1,537 healthy individuals from multiple cohorts, whereas the test set included 2 groups, 99 traumatic brain injury (TBI) patients and 113 healthy controls, all scanned on the same scanner. (A) Conventional Statistical Parametric Mapping (SPM) structural preprocessing pipeline was used to generate gray and white matter maps, normalized to Montreal Neurological Institute (MNI) space and modulated to retain data relating to brain size. (B) Separately for gray and white matter, all 1,749 data sets were converted to a kernel matrix based on voxelwise similarity using Pronto. (C) The training data only were run through a supervised learning stage where a Gaussian Processes Regression (GPR) machine was trained to recognize patterns of imaging data that matched a given age label. To assess model accuracy, 10-fold cross-validation was con- ducted where 10% of samples were excluded from the training step and the ages of these samples were estimated. This was iterated 9 further times to generate age predictions on all samples. (D) The trained GPR model was then applied to the 2 test data sets, to assess accuracy of the model on healthy controls and then predict brain age of TBI patients. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.] 574 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI FIGURE 3: Machine learning model provides accurate age prediction in healthy training set. Predicted age for each healthy individual in the training set (n5 1,537) is shown, derived by running 10-fold cross-validation on the Gaussian Processes Regression model. (A) Chronological age (x-axis) is plotted against predicted age (y-axis), for gray matter (dark gray circles). (B) Chronological age and predicted age for white matter (light gray triangles). Diagonal dashed line represents the line of iden- tity (x5 y). testing resulted in a corrected p-value of 0.001 for both Statistical Analysis Model accuracy was assessed using Pearson correlation coeffi- GM and WM models. The mean PAD for the training cients between predicted age and chronological age of the train- group was 20.003 years (67.82) for GM and 0.037 ing subjects. The proportion of the variation explained by the years (67.74) for WM. For the healthy control test data trained model (R ) was examined, as was the mean absolute set, age was accurately predicted for both GM error (MAE) between predicted and chronological age. Accuracy 2 (r5 0.931, R 5 0.867, MAE5 5.80) and WM was also evaluated for the control test data set, using the above (r5 0.931, R 5 0.867, MAE5 6.35). These results vali- parameters. Predicted age was subtracted from chronological dated the generalizability of the GPR modeling age, generating a predicted age difference (PAD) score per par- approach, as high accuracy was achieved when training ticipant. PAD scores were statistically analyzed in R (http:// the model on one data set and testing on an entirely www.R-project.org/) using an analysis of covariance (ANCOVA) independent data set. to test for group differences between TBI patients and controls, while covarying for age and sex. Within the TBI group, PAD TBI Patients Have Increased Brain Age was also assessed for Spearman nonparametric correlations with TBI patients (GM mean PAD5 4.666 10.8; WM mean age, TSI, neuropsychological measures, and differences between PAD5 5.976 11.22) showed increased PAD compared subgroups based on TBI severity, mechanism of injury, and the to controls (GM mean PAD5 0.076 7.41; WM mean presence of contusions. Analysis of PAD score was repeated for both GM and WM, and an exploratory analysis of the relation- PAD5 2.066 7.41; Fig 4). Comparing these PAD ship between the 2 brain tissue classes was also carried out. scores between TBI and control groups showed signifi- cant differences for both GM (F5 14.8, p< 0.001) and Results WM (F5 9.7, p5 0.002). The discrepancy between predicted and chronologi- Chronological Age Can Be Predicted from cal age was related to the severity of injury (Fig 5A, B). Structural Neuroimaging The model was able to accurately predict chronological Patients with a moderate/severe classification showed age for both training and control test data sets. For the increased mean PAD score (GM mean5 5.726 10.97; training set (Fig 3), age was accurately predicted based WM mean5 7.246 11.64), whereas mild (probable) on both GM (predicted–chronological age correlation patients were not significantly different from controls r5 0.921, R 5 0.848, MAE5 6.2 years) and WM (GM mean520.426 8.48; WM mean520.146 6.1). (r5 0.922, R 5 0.851, MAE5 6.16 years). Permutation The presence of focal lesions had no influence on PAD April 2015 575 ANNALS of Neurology FIGURE 4: Traumatic brain injury (TBI) patients show increased predicted age difference (PAD) score for gray matter and white matter. Boxplots of PAD score are shown, calculated by subtracting chronological age from predicted age for the test data sets of 99 TBI patients and 113 healthy controls. (A) PAD scores derived from the gray matter model showing a significant increase in TBI patients. (B) PAD scores from white matter also show an increase in TBI patients. score, as there were no significant differences between an increasing discrepancy between predicted and chrono- those with and without contusions for either GM (mean: logical age with a longer TSI (Fig 6). As there was also a 5.99 vs 5.55, p5 0.85) or WM (mean: 6.89 vs. 7.44, correlation between TSI and age (r5 0.28, p5 0.0047) p5 0.83). Furthermore, when limiting the TBI patient and between PAD score and age (GM: r5 0.18, group to only moderate/severe lesion-free patients p5 0.068; WM: r5 0.21, p5 0.037), a partial correla- (n5 31), ANCOVA for GM and WM still showed sig- tion approach was used to examine directly the relation- nificantly greater PAD scores compared to controls ship between PAD score and time since TBI, to remove (p< 0.001). Although mean PAD score for TBI patients variance associated with age. Three outliers in terms of and controls was greater for WM than GM, the relatively TSI were present in the TBI group (168, 206, and 563 high variability meant that this difference was not signifi- months), defined as being 62 SD away from the mean cant (p> 0.1). (mean5 28.386 63.58, 2-SD range5 0–155.54 months). These patients were not driving the association, Brain Atrophy Correlates with Cognitive as the results were still highly significant after these Impairment patients were removed (GM: r5 0.506, p< 0.001; WM: Performance on a number of cognitive measures corre- r5 0.467, p< 0.001). Removing patients with focal lated strongly with PAD scores (Table 2). After correc- lesions also did not alter the association between TSI and tion for the multiple tests conducted (using the false PAD score, as analysis in contusion-free patients (n5 47) discovery rate ), measures relating to information proc- was still significant (GM: r5 0.353, p5 0.01; WM: essing speed and memory were significantly correlated r5 0.342, p5 0.016). with GM and WM PAD, with increased reaction time being related to increases in PAD score. Executive func- Brain "Ageing" Does Not Vary in Patients with tion measures were significantly correlated for GM PAD Different Mechanism of Injury scores, but not for WM. Scores on the WASI subscales We assessed whether the mechanism of injury affected were not correlated with PAD scores. brain ageing. Patients who suffered TBI due to RTA, assault, or fall all showed increased mean PAD scores Evidence for Accelerated Atrophy following TBI There was a strong correlation between PAD score and (5.93, 6.15, and 4.11 for GM; 7.52, 4.42, and 6.51 for TSI for both GM (r5 0.535, p< 0.001) and WM WM), indicating that TBI patients’ brains appeared to (r5 0.496, p< 0.001), controlling for age. This reflected be older, irrespective of the cause of injury (see Fig 5C, 576 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI FIGURE 5: Gray matter (GM) and white matter (WM) predicted age difference (PAD) score, stratified by injury severity and injury mechanism. Boxplots of PAD score in the traumatic brain injury (TBI) patient group are shown, stratified by clinical char- acteristics. (A) GM PAD score distributions for each Mayo classification: probable/mild, moderate/severe, indicating that brain age is only increased in moderate/severe patients, not in mild TBI. (B) Mayo classification for WM. (C) GM PAD score by mech- anism of injury (assault, fall, road–traffic accident [RTA]), indicating that similar levels of increased brain ageing occur independ- ent of mechanism of injury. (D) Mechanism of injury for WM. D). There were no significant differences in PAD on average >4 years older than the patient’s chronologi- between mechanism subgroups for either GM or WM. cal age. This discrepancy was only seen in patients with The correlation between TSI and PAD was still present more severe injuries, was independent of the mechanism for patients injured during RTAs and assaults, although of injury, and was predictive of cognitive impairment. not for sufferers of falls. Mean TSI associated with falls These results support the theory proposed by Moretti was only 19.8 (635.2) months, considerably less than and colleagues that TBI may hasten the ageing process for RTA (33.26 43.1 months) and assaults and is in keeping with long-term structural and func- (41.56 120.7 months). This decreased duration and nar- tional brain abnormalities reported in neuroimaging and 11–13,25,26 rower distribution may have led to lower sensitivity to neuropathology studies of TBI patients. the effects of TSI on PAD in sufferers of falls. Animal work demonstrates a progressive loss of brain tissue after experimental TBI. A number of mech- Discussion anisms for this have been proposed, including chronic 12,26 Using a multivariate method to investigate the spatial neuroinflammation, Wallerian degeneration of patterns of age-associated brain atrophy, we show that WM, and the deposition of abnormal tau and TBI produces a pattern of structural brain changes that amyloid-b proteins. These processes lead to alterations affects the apparent brain age of both GM and WM. in cellular morphology, loss of trophic support, and even- The predictive model we generated was highly accurate tual cell death and gross atrophy. It is also possible that at estimating chronological age in healthy participants, a reduction in brain volume might be due to the clear- based only on the appearance of T1 MRI scans. In con- ance of tissue damaged at the time of injury. Previous trast, following TBI, the model estimated brain age to be neuroimaging studies have typically shown atrophy when April 2015 577 ANNALS of Neurology TABLE 2. Relationship between Neuropsychological Measures and PAD in TBI Patients Cognitive Neuropsychological Test No. Mean GM PAD, p WM PAD, p Domain rho rho a a Processing Trail Making Test A, s 90 29.59 (12.15) 0.338 0.001 0.379 <0.001 speed a a Trail Making Test B, s 90 68.79 (39.79) 0.343 0.001 0.271 0.009 a a Stroop color naming, s 90 34.47 (8.83) 0.279 0.007 0.281 0.006 a a Stroop word reading, s 90 23.93 (5.49) 0.269 0.009 0.306 0.003 Choice reaction task 66 478 (124) 0.331 0.005 0.243 0.047 median reaction time, ms Executive Trail Making 90 39.15 (32.51) 0.262 0.011 0.154 0.146 function Test B minus A, s Inhibition/switching, s 89 69.54 (22.18) 0.247 0.018 0.205 0.053 Inhibition/ 89 37.24 (18.41) 0.213 0.043 0.168 0.115 switching minus baseline Stroop performance, s Letter fluency total 89 39.01 (12.73) 20.271 0.009 20.133 0.214 Intellectual WASI similarities 90 37.04 (5.22) 20.160 0.132 20.136 0.200 ability WASI matrix reasoning 88 26.34 (5.53) 20.104 0.336 0.013 0.908 a a Memory People Test 90 22.94 (7.33) 20.253 0.015 20.254 0.014 immediate recall Correlations with PAD score were conducted with variance accounting for chronological age partialed out, using the Spearman rank-order approach. Denotes statistical significance after false discovery rate correction for multiple comparisons. GM5 gray matter; PAD5 predicted age difference score; TBI5 traumatic brain injury; WASI5 Wechsler Abbreviated Scale for Intelligence; WM5 white matter. comparing acute and chronic scans, making it difficult to age. Discrepancies between brain and chronological age disentangle these possibilities. One longitudinal study appear to be biologically informative, perhaps reflecting has shown that GM volume decreases continue between important differences in susceptibility or resistance to 1 and 4 years after TBI, demonstrating that tissue loss age-related pathology. This possibility is supported by continues well beyond the stage when clearance of the relationship we observed between PAD score and acutely damaged tissue is likely to have finished. This cognitive function. Our structural imaging measures cor- suggests a progressive process, which is supported by our related with cognitive impairment in domains typically 10,11 observation that patients who were assessed further from affected by TBI, namely information processing their injury (high TSI) showed more atrophy. This sup- speed, memory, and executive function. As the vast ports the idea that tissue loss accelerates over time, in majority of these patients were at least 3 months postin- keeping with a progressive neurodegenerative process jury, it is likely that such impairments are persistent, triggered by the injury (see Figs 1 and 4 and reviews by rather than acute and transient, implying that there is a 2 9 Smith and colleagues and Bigler ). relationship between the observed brain changes and the Our model explained much of the variation in chronic post-TBI cognitive profile. Intriguingly, these are chronological age, in line with previous research showing the same cognitive domains often affected by age-related that multivariate brain imaging analysis can accurately cognitive decline, which leads us to speculate that PAD 15,30 predict chronological age. Much like outwardly visi- score may be relevant to the effects of ageing in a ble signs of age such as the presence of wrinkles or gray broader sense. In particular, the strongest associations hair, this shows that the brain also varies in its apparent with our measure of brain ageing were found with 578 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI FIGURE 6: Gray matter and white matter predicted age difference (PAD) score increases with greater time since injury (TSI). Scatterplots depicting the relationship between PAD score (x-axis) and TSI (y-axis) are shown. Plotted values are the residuals derived from a linear regression with PAD score or TSI, regressing out chronological age. (A) Gray matter (dark gray circles) PAD scores, with dashed lines representing the locally weighted scatterplot smoothing (lowess) line calculated (dashed gray line). B) White matter (light gray triangles) PAD scores with lowess line (dashed light gray line). Both analy- ses were conducted after the removal of 3 outliers, identified based on having a TSI of 62 standard deviations from the mean. information processing speed measures, and impairments phy is important, as it suggests that the underlying in this domain have been posited as central to the cogni- differences in brain structure are unlikely to be secondary tive decline associated with typical ageing. The nature to nonspecific neurological or demographic factors that of our sample did not permit an exhaustive examination might be expected to vary across individuals with TBI of TBI-related and age-related cognitive dysfunction; from very different causes. For example, a possible con- however, our interpretation of results supports the idea found is that patients predisposed to TBI might have that the pattern of brain changes detected in TBI smaller brains prior to injury. However, there is no evi- patients has similar functional consequences to normal dence from other studies that this is the case, and it is ageing, albeit occurring in an accelerated form. Measures highly unlikely that patients with such contrasting mech- of age discrepancy might be useful for screening clinical anisms of injury would also show the same systematic and population samples to identify those at increased bias in brain size. risk of age-associated pathologies, for quantifying the It is unclear whether the brain atrophy we observe effects of vascular risk factors or neuropsychiatric diseases after TBI reflects ongoing neurodegeneration triggered by on general brain health, and for stratifying patients for the injury or an interaction with normal ageing. In part, targeted treatments or clinical trial enrollment. this reflects the complexity of defining ageing. Rates of As predicted, the degree of apparent ageing reflects ageing vary between individuals, but also can affect dif- the severity of initial injury. In contrast to moderate/ ferent tissues in the same person at different rates. This severe TBI, patients with minor TBI showed no brain makes comprehensively modeling age challenging. One atrophy. This suggests that a significant biomechanical limit of the work is that we focus only on a single facet force is necessary to trigger ongoing neurodegenerative of ageing, demonstrating that individuals have more processes that lead to progressive atrophy. In addition, brain atrophy after TBI. We are unable to disentangle the mechanism of injury did not influence PAD score, the degree to which this atrophy results from an accelera- with those suffering falls, assaults, and RTAs showing tion of processes seen in healthy ageing, whether the similar levels of atrophy. That widely varying mecha- injury triggers new neurodegenerative processes, or the nisms of TBI result in similar patterns of premature atro- extent to which these two possibilities are inter-related. April 2015 579 ANNALS of Neurology Processes involved in neurodegeneration such as inflam- that resemble the atrophy seen during ageing. The accel- mation and the accumulation of misfolded proteins are erated trajectory of brain atrophy we observed is in keep- also seen to varying degrees in healthy ageing, and com- ing with a progressive neurodegenerative process parative studies focusing on these factors will be neces- triggered by the injury. The effects of age and injury are sary to define to what extent TBI patients resemble likely to act synergistically, leading to deficits in informa- healthy older individuals or should be viewed as having tion processing speed and other neuropsychological meas- distinct neuropathology. ures known to be impaired in normal ageing. Future A number of other potential limitations of the studies could use PAD score as a predictor of clinical study should also be noted. First, some TBI patients outcome or as a surrogate marker of treatment efficacy. have small focal lesions that could potentially cause prob- The work gives insight into the etiology of the long-term lems with the image registration techniques that our age chronic effects of TBI and has implications for the long- prediction requires. However, we use advanced registra- term care and potential future treatments for TBI tion algorithms that perform well in our patient group, patients, as methods that attenuate the negative effects of and our results are similar when the analysis is restricted ageing may also be effective treatments for patients who to patients without focal lesions. Second, the estimates of have suffered a TBI. age were variable, indicating a degree of residual mea- surement error, which is to be expected as we were using Acknowledgment only brain structure to predict chronological age. Never- This research was supported by a European Union Sev- theless, the large sample size meant we were sufficiently enth Framework Programme grant to the Comorbidity powered to detect statistically significant group differen- in Relation to AIDS (COBRA) project (FP-7-HEALTH ces in PAD score. 305522; J.H.C.), an National Institute for Health We exploited the wide range of TSI to explore the Research (NIHR) Professorship (NIHR-RP-011-048; chronic effects of TBI. Our analysis indicated both that D.J.S.), and the NIHR Imperial Biomedical Research increased atrophy is present after TBI, and also that the Centre. amount of atrophy increases over time. Removing the Data used in the preparation of this article were patients with the greatest TSI did not alter the results, obtained from the Alzheimer’s Disease Neuroimaging implying that this apparent acceleration is a robust effect Initiative (ADNI) database (http://adni.loni.usc.edu/). As even over a relatively short time period. As there are a such, the investigators within ADNI contributed to the number of potential confounds inherent with cross- design and implementation of ADNI and/or provided sectional analyses, our evidence for accelerating atrophy data but did not participate in analysis or writing of this should be interpreted with caution. For example, brain report. The draft manuscript was certified by the ADNI so ageing is likely to be in part genetically mediated, Data and Publications Committee as meeting the ADNI information about genotype should be added in future data sharing requirements. Full details of the ADNI data studies to potentially explain even greater amounts of sharing policy and further information fully acknowledg- variance in age. In addition, the severity of the initial ing the relevant funding sources can be found here: injury will influence the extent of brain atrophy. We only http://adni.loni.usc.edu/wp-content/uploads/how_to_ observed increased atrophy in patients with moderate/ apply/ADNI_DSP_Policy.pdf severe injuries, which shows the expected impact of The views expressed are those of the authors and not injury on apparent brain age. However, within the mod- necessarily those of the National Health Service, the erate/severe group there will be variations in injury sever- NIHR, or the Department of Health. ity, and these could potentially confound the relationship between TSI and brain atrophy. Individual indices of injury severity such as Glasgow Coma Scale score and duration of post-traumatic amnesia are known to be Potential Conflicts of Interest crude measures of injury severity. Therefore, large lon- D.J.S.: investigator led grant, Pfizer. gitudinal studies will be necessary to completely resolve these issues, ideally with individual follow-up at multiple time points to allow longitudinal within-subject analysis that incorporates other genetic and biomarker informa- References tion about the ageing process. 1. Smith DH, Chen XH, Pierce JES, et al. Progressive atrophy and In summary, our study is the first empirical in vivo neuron death for one year following brain trauma in the rat. demonstration that TBI causes structural brain changes J Neurotrauma 1997;14:715–727. 580 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI 2. Smith DH, Johnson VE, Stewart W. Chronic neuropathologies of 19. Koutsouleris N, Davatzikos C, Borgwardt S, et al. Accelerated single and repetitive TBI: Substrates of dementia? Nat Rev Neurol brain aging in schizophrenia and beyond: a neuroanatomical 2013;9:211–221. marker of psychiatric disorders. Schizophr Bull 2014;40:1140– 3. Moretti L, Cristofori I, Weaver SM, et al. Cognitive decline in older adults with a history of traumatic brain injury. Lancet Neurol 2012; 20. Salthouse TA. The processing-speed theory of adult age differen- 11:1103–1112. ces in cognition. Psychol Rev 1996;103:403–428. 4. Bazarian JJ, Cernak I, Noble-Haeusslein L, et al. Long-term neuro- 21. Ashburner J. A fast diffeomorphic image registration algorithm. logic outcomes after traumatic brain injury. J Head Trauma Reha- Neuroimage 2007;38:95–113. bil 2009;24:439–451. 22. Schrouff J, Rosa MJ, Rondina JM, et al. PRoNTo: pattern recogni- 5. Hedden T, Gabrieli JDE. Insights into the ageing mind: a view tion for neuroimaging toolbox. Neuroinformatics 2013;11:319– from cognitive neuroscience. Nat Rev Neurosci 2004;5:87–96. 337. 6. Ferreira D, Molina Y, Machado A, et al. Cognitive decline is medi- 23. Doyle OM, Ashburner J, Zelaya FO, et al. Multivariate decoding ated by gray matter changes during middle age. Neurobiol Aging of brain images using ordinal regression. Neuroimage 2013;81: 2014;35:1086–1094. 347–357. 7. Sowell ER, Peterson BS, Thompson PM, et al. Mapping cortical 24. Benjamini Y, Hochberg Y. Controlling the false discovery rate—a change across the human life span. Nat Neurosci 2003;6:309–315. practical and powerful approach to multiple testing. J R Stat Soc B Met 1995;57:289–300. 8. Raz N, Rodrigue KM. Differential aging of the brain: patterns, cog- nitive correlates and modifiers. Neurosci Biobehav Rev 2006;30: 25. Povlishock JT, Katz DI. Update of neuropathology and neurologi- 730–748. cal recovery after traumatic brain injury. J Head Trauma Rehabil 2005;20:76–94. 9. Bigler ED. Traumatic brain injury, neuroimaging, and neurodegen- eration. Front Hum Neurosci 2013;7:395. 26. Johnson VE, Stewart JE, Begbie FD, et al. Inflammation and white matter degeneration persist for years after a single traumatic brain 10. Draper K, Ponsford J. Cognitive functioning ten years following injury. Brain 2013;136:28–42. traumatic brain injury and rehabilitation. Neuropsychology 2008; 22:618–625. 27. Meythaler JM, Peduzzi JD, Eleftheriou E, Novack TA. Current con- cepts: diffuse axonal injury-associated traumatic brain injury. Arch 11. Kinnunen KM, Greenwood R, Powell JH, et al. White matter dam- Phys Med Rehabil 2001;82:1461–1471. age and cognitive impairment after traumatic brain injury. Brain 2011;134:449–463. 28. Johnson VE, Stewart W, Smith DH. Widespread tau and amyloid- beta pathology many years after a single traumatic brain injury in 12. Ramlackhansingh AF, Brooks DJ, Greenwood RJ, et al. Inflamma- humans. Brain Pathol 2012;22:142–149. tion after trauma: microglial activation and traumatic brain injury. Ann Neurol 2011;70:374–383. 29. Farbota KD, Sodhi A, Bendlin BB, et al. Longitudinal volumetric changes following traumatic brain injury: a tensor-based mor- 13. Sharp DJ, Beckmann CF, Greenwood R, et al. Default mode net- phometry study. J Int Neuropsychol Soc 2012;18:1006–1018. work functional and structural connectivity after traumatic brain injury. Brain 2011;134:2233–2247. 30. Groves AR, Smith SM, Fjell AM, et al. Benefits of multi-modal fusion analysis on a large-scale dataset: life-span patterns of inter- 14. Ross DE. Review of longitudinal studies of MRI brain volumetry in subject variability in cortical morphometry and white matter micro- patients with traumatic brain injury. Brain Inj 2011;25:1271–1278. structure. Neuroimage 2012;63:365–380. 15. Franke K, Ziegler G, Kloppel € S, Gaser C. Estimating the age of 31. Simm A, Johnson TE. Biomarkers of ageing: a challenge for the healthy subjects from T1-weighted MRI scans using kernel meth- future. Exp Gerontol 2010;45:731–732. ods: exploring the influence of various parameters. Neuroimage 2010;50:883–892. 32. Filippini N, Ebmeier KP, MacIntosh BJ, et al. Differential effects of the APOE genotype on brain function across the lifespan. Neuro- 16. Brown T, Kuperman JM, Chung Y, et al. Neuroanatomical assess- image 2011;54:602–610. ment of biological maturity. Curr Biol 2012;22:1693–1698. 33. Malec JF, Brown AW, Leibson CL, et al. The mayo classification 17. Dosenbach NUF, Nardos B, Cohen AL, et al. Prediction of individ- system for traumatic brain injury severity. J Neurotrauma 2007;24: ual brain maturity using fMRI. Science 2010;329:1358–1361. 1417–1424. 18. Gaser C, Franke K, Kloppel € S, et al. BrainAGE in mild cognitive impaired patients: predicting the conversion to Alzheimer’s dis- ease. PLoS One 2013;8:e67346. April 2015 581 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Annals of Neurology Pubmed Central

Prediction of brain age suggests accelerated atrophy after traumatic brain injury

Cole, James H; Leech, Robert; Sharp, David J; ,
Annals of Neurology , Volume 77 (4) – Mar 25, 2015
Download PDF
11 pages

Loading next page...
 
/lp/pubmed-central/prediction-of-brain-age-suggests-accelerated-atrophy-after-traumatic-UccE89g12B

References (35)

Publisher
Pubmed Central
Copyright
© 2015 The Authors Annals of Neurology published by Wiley Periodicals, Inc. on behalf of American Neurological Association.
ISSN
0364-5134
eISSN
1531-8249
DOI
10.1002/ana.24367
Publisher site
See Article on Publisher Site

Abstract

RESEARCH ARTICLE Prediction of Brain Age Suggests Accelerated Atrophy after Traumatic Brain Injury James H. Cole, PhD, Robert Leech, PhD, and David J. Sharp, PhD, for the Alzheimer’s Disease Neuroimaging Initiative Objective: The long-term effects of traumatic brain injury (TBI) can resemble observed in normal ageing, suggesting that TBI may accelerate the ageing process. We investigate this using a neuroimaging model that predicts brain age in healthy individuals and then apply it to TBI patients. We define individuals’ differences in chronological and predicted structural "brain age," and test whether TBI produces progressive atrophy and how this relates to cognitive function. Methods: A predictive model of normal ageing was defined using machine learning in 1,537 healthy individuals, based on magnetic resonance imaging–derived estimates of gray matter (GM) and white matter (WM). This ageing model was then applied to test 99 TBI patients and 113 healthy controls to estimate brain age. Results: The initial model accurately predicted age in healthy individuals (r5 0.92). TBI brains were estimated to be "older," with a mean predicted age difference (PAD) between chronological and estimated brain age of 4.66 years (610.8) for GM and 5.97 years (611.22) for WM. This PAD predicted cognitive impairment and correlated strongly with the time since TBI, indicating that brain tissue loss increases throughout the chronic postinjury phase. Interpretation: TBI patients’ brains were estimated to be older than their chronological age. This discrepancy increases with time since injury, suggesting that TBI accelerates the rate of brain atrophy. This may be an important factor in the increased susceptibility in TBI patients for dementia and other age-associated conditions, motivating fur- ther research into the age-like effects of brain injury and other neurological diseases. ANN NEUROL 2015;77:571–581 7,8 raumatic brain injury (TBI) causes long-term struc- cognitive impairment, and brain volume loss. Insults, Ttural and functional alterations to the brain. Some of such as TBI, may trigger a sequence of neurobiological 1,2 these changes are thought to be progressive in nature, events that alter that trajectory, prematurely causing brain and potentially underlie the increased risk for early cogni- atrophy, and potentially manifesting as an early onset of 3 4 9 tive decline and dementia observed in TBI patients. Sim- neurodegeneration. As illustrated in Figure 1, an envi- ilar behavioral and anatomical changes are also associated ronmental insult like TBI might cause a one-off increase 5,6 with normal ageing, raising the possibility that the in apparent "brain age," or could result in an ongoing chronic consequences of TBI may contribute to the prema- interaction between injury and ageing-related or other ture development of age-associated changes to the brain. neurodegenerative processes that cause progressive brain Normal ageing can be considered as the progression atrophy. In the latter case, as more time passes since the along a temporal trajectory, where individuals gradually TBI occurred, the greater the discrepancy between chro- accumulate pathologies associated with physical decline, nological age and estimated brain age will be. This A complete listing of Alzheimer’s Disease Neuroimaging Initiative investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf View this article online at wileyonlinelibrary.com. DOI: 10.1002/ana.24367 Received Aug 21, 2014, and in revised form Jan 6, 2015. Accepted for publication Jan 19, 2015. Address correspondence to Dr Cole, Burlington Danes Building, Hammersmith Hospital, Imperial College London, London, W12 0NN, United Kingdom. E-mail: [email protected] From the and Computational, Clinical, and Cognitive Neuroimaging Laboratory, Department of Medicine, Imperial College London, London, United Kingdom Additional Supporting Information may be found in the online version of this article. V 2015 The Authors Annals of Neurology published by Wiley Periodicals, Inc. on behalf of American Neurological Association. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. 571 ANNALS of Neurology teristic changes to cognition seen during normal ageing affect the domains of executive function, memory, and information processing speed. TBI patients show a simi- lar pattern of cognitive deficits, further suggesting links between TBI and ageing of the brain. Here we developed and tested a predictive model of brain age. Machine learning techniques were used to define a model that accurately predicted chronological age in healthy individuals. The model was then applied to brain images from TBI patients, allowing a prediction of their brains’ ages to be made. It was expected that TBI patients’ brains would be older than their chronological age and that this discrep- ancy would increase with longer time since injury (TSI), FIGURE 1: Model of premature brain ageing in traumatic brain reflecting a progressive atrophy of brain tissue. Furthermore, injury. Illustration of the conceptual framework for the investi- gation of brain age in traumatic brain injury (TBI). The short- we hypothesized that the discrepancy between chronological dashed line represents the trajectory of healthy ageing as age age and predicted brain age would be reflected in cognitive (x-axis) increases, against a background gradient of increasing changes that would be consistent with age-related cognitive susceptibility to age-related pathology (y-axis), such as cogni- tive decline and dementia. Occurrence of TBI is indicated (black impairment normally seen in older individuals. arrow), with acute pathology causing an immediate departure from a healthy brain state. Two alternative brain ageing trajec- Subjects and Methods tories post-TBI are shown. The long-dashed "additive effects" line depicts a trajectory assuming a one-off hit, with damage Participants leading to the patient’s brain structure resembling an older individual, followed by a normal rate of subsequent ageing. TRAINING SET. T1-weighted magnetic resonance imaging The dash–dot "interactive effects" line represents an acceler- (MRI) data from 1,537 healthy controls were obtained from 8 ated rate of brain atrophy caused by TBI and an interaction publically accessible neuroimaging initiatives (Supplementary with normal ageing processes, with the discrepancy between Table). This provided an unbiased source of data with which to normal ageing and pathological changes increasing the greater the time since injury (TSI). Comparing predicted age difference train the age prediction model that was entirely independent (PAD) scores (i; dashed black line) and (ii; solid black line)illus- from the TBI and control test data sets. Controls had no his- trates how a greater PAD score would be expected under the tory of significant neurological or psychiatric problems, with interactive effects model with accelerating atrophy (i), com- further specific recruitment criteria made by each independent pared to the added effects model (ii), at equivalent TSI (figure study. Exclusions were made due to poor data quality leading adapted from Smith and colleagues ). [Color figure can be viewed in the online issue, which is available at wileyonline to image processing failure, identified during imaging quality library.com.] assessment. All training set data had been previously anony- mized, and ethical approval and informed consent were obtained by each specific study. possibility is consistent with the progressive decline asso- ciated with TBI, even years after injury, as demonstrated TEST SET. Ninety-nine patients with persistent neurological 10 11–14 1 by neuropsychological, neuroimaging, and animal problems after TBI (72 males, mean age6 standard deviation research. [SD]: 37.986 12.43 years) were recruited (Table 1). A comparison group of 113 healthy controls assessed on the same scanner was Using neuroimaging, it is possible to predict age in used to validate the accuracy of the age prediction model (49 healthy individuals, allowing the discrepancy between males, 43.36 20.24 years). All patients were scanned at least 1 chronological age and predicted brain age to be calculated. month post-injury (mean5 28.4 months), with a range of 1 to In clinical samples, this discrepancy can be considered as 563 months. This variation allowed us to examine the influence of an index of the deviation from a normal ageing trajectory. TSI on brain changes. The severity of TBI was classified using the Estimates of brain age, derived using machine learning, Mayo Classification System criteria, with 17% being classified as have previously been used in a number of contexts. These mild (probable) and 83% being moderate/severe. Cause of injury include measuring normal brain maturation during devel- was reported as follows: road–traffic accident (RTA; 39%), fall 16,17 opment, predicting conversion from mild cognitive (24%), assault (23%), sport-related injury (6%), other (7%). Cere- impairment to Alzheimer disease, and in neurodevelop- bral contusions were identified in 52 patients (53%) by an experi- mental disorders such as schizophrenia and borderline per- enced neuroradiologist. Exclusion criteria were as follows: sonality disorder, where patients were shown to have psychiatric or neurological illness, previous traumatic brain injury, antiepileptic medication, current or previous drug or alcohol abuse, apparently "older" brains. The deviation from normal MRI contraindication. All participants gave written informed con- ageing may reflect important neurological changes relating sent, and the local ethics committee approved the study. to clinical features such as cognitive impairment. Charac- 572 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI TABLE 1. Details of TBI Patients Characteristic TBI Patients Controls No. 99 113 Age, yr 37.986 12.43 43.36 20.24 Sex, M/F 72/27 49/64 Time since TBI, mo 28.46 63.57 — Mechanism of injury, RTA/fall/assault/sports/other 39/24/23/6/7 — Mayo Clinic criteria, probable–mild/moderate–severe 17/82 — Patient has contusions on MRI, present/absent 52/47 — GCS 9.716 4.75 — Patient medicated at first visit, medicated/not medicated 55/44 — Values are reported in mean6 standard deviation or absolute numbers. F5 female; GCS5 Glasgow Coma Scale; M5 male; MRI5 magnetic resonance imaging; RTA5 road–traffic accident; TBI5 traumatic brain injury. MACHINE LEARNING PREDICTION OF AGE. Machine Neuropsychological testing was carried out using a stand- learning analysis (see Fig 2B) was conducted using the Pattern ardized battery of tests previously shown to be sensitive to cog- Recognition for Neuroimaging Toolbox (Pronto ) and run on nitive abnormalities in TBI. These were the Trail Making GM and WM separately. Data from all subjects were converted Task, Stroop color naming and word reading, Wechsler Abbre- to a similarity matrix kernel to improve computational effi- viated Scale for Intelligence (WASI) matrix reasoning and simi- ciency in Pronto by generating a vector representation of voxel- larities subscales, Choice Reaction task, People and Doors wise intensity levels for all data and calculating the dot immediate recall test, letter fluency, and task inhibition/ product between each image. Next, the data were mean- switching. centered and a Gaussian Processes Regression (GPR) model was Procedures defined using age as the dependent variable and the similarity A high-level overview of the methods is provided in Figure 2. matrix of imaging data as the independent variables. GPR is a Methods are summarized below. machine learning extension of the classical regression model, which incorporates nonlinear and Gaussian probabilistic ele- MRI ACQUISITION PARAMETERS. For the training set of ments to allow quantitative predictions to be made using con- high-resolution T1-weighted magnetic resonance images, multi- tinuous variables. ple scanners were used, including different vendors, field strengths, and acquisition protocols (see Supplementary Table). MODEL VALIDATION. Model validation (see Fig 2C) pro- For the test set, T1 images were acquired using a Philips 3T ceeded in 3 stages to ensure independence between training and Intera scanner (Philips Medical Systems, Best, the Netherlands) test sets and to enable an unbiased demonstration of model gener- with the following parameters: matrix size5 208 3 208, slice alizability. The first stage involved running 10-fold cross-validation thickness5 1.2mm, 0.94 x 0.94mm in-plane resolution, 150 on the training set, to determine the accuracy of the model. This slices, repetition time5 9.6 milliseconds, echo time5 4.5 milli- entailed randomly selecting one-tenth of the training data to be a seconds, flip angle5 8 . temporary test set, with the remainder used for model definition. Using the learned pattern from this training set, age was predicted IMAGE PREPROCESSING. All MRI data were preprocessed on the temporary test set. This process was iterated until all images (see Fig 2A) using the Statistical Parametric Mapping (SPM8) had been included in the test set and had an age value predicted. software package (University College London, London, UK). This was followed by permutation testing with 1,000 randomiza- This included tissue segmentation into gray matter (GM) and tions to derive a p-value for each model. The second stage involved white matter (WM) maps, then registration using the nonlinear validating the model on the independent controls test data set. DARTEL algorithm to Montreal Neurological Institute space Here, the model was defined using the entire training set and resampling with a 4mm smoothing kernel. Each tissue class (N5 1,537) and then used to predict the brain ages for the control (ie, GM and WM) was processed independently after segmenta- test data set. Finally, the trained model was applied to estimate pre- tion. The preprocessing procedure ensured that all images were dicted brain ages for the TBI patients (see Fig 2D). well aligned and appropriate for voxelwise analysis at the machine learning stage. April 2015 573 ANNALS of Neurology FIGURE 2: Overview of the study methods. Study data comprised 2 sets, training and test. The training set used structural magnetic resonance imaging from 1,537 healthy individuals from multiple cohorts, whereas the test set included 2 groups, 99 traumatic brain injury (TBI) patients and 113 healthy controls, all scanned on the same scanner. (A) Conventional Statistical Parametric Mapping (SPM) structural preprocessing pipeline was used to generate gray and white matter maps, normalized to Montreal Neurological Institute (MNI) space and modulated to retain data relating to brain size. (B) Separately for gray and white matter, all 1,749 data sets were converted to a kernel matrix based on voxelwise similarity using Pronto. (C) The training data only were run through a supervised learning stage where a Gaussian Processes Regression (GPR) machine was trained to recognize patterns of imaging data that matched a given age label. To assess model accuracy, 10-fold cross-validation was con- ducted where 10% of samples were excluded from the training step and the ages of these samples were estimated. This was iterated 9 further times to generate age predictions on all samples. (D) The trained GPR model was then applied to the 2 test data sets, to assess accuracy of the model on healthy controls and then predict brain age of TBI patients. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.] 574 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI FIGURE 3: Machine learning model provides accurate age prediction in healthy training set. Predicted age for each healthy individual in the training set (n5 1,537) is shown, derived by running 10-fold cross-validation on the Gaussian Processes Regression model. (A) Chronological age (x-axis) is plotted against predicted age (y-axis), for gray matter (dark gray circles). (B) Chronological age and predicted age for white matter (light gray triangles). Diagonal dashed line represents the line of iden- tity (x5 y). testing resulted in a corrected p-value of 0.001 for both Statistical Analysis Model accuracy was assessed using Pearson correlation coeffi- GM and WM models. The mean PAD for the training cients between predicted age and chronological age of the train- group was 20.003 years (67.82) for GM and 0.037 ing subjects. The proportion of the variation explained by the years (67.74) for WM. For the healthy control test data trained model (R ) was examined, as was the mean absolute set, age was accurately predicted for both GM error (MAE) between predicted and chronological age. Accuracy 2 (r5 0.931, R 5 0.867, MAE5 5.80) and WM was also evaluated for the control test data set, using the above (r5 0.931, R 5 0.867, MAE5 6.35). These results vali- parameters. Predicted age was subtracted from chronological dated the generalizability of the GPR modeling age, generating a predicted age difference (PAD) score per par- approach, as high accuracy was achieved when training ticipant. PAD scores were statistically analyzed in R (http:// the model on one data set and testing on an entirely www.R-project.org/) using an analysis of covariance (ANCOVA) independent data set. to test for group differences between TBI patients and controls, while covarying for age and sex. Within the TBI group, PAD TBI Patients Have Increased Brain Age was also assessed for Spearman nonparametric correlations with TBI patients (GM mean PAD5 4.666 10.8; WM mean age, TSI, neuropsychological measures, and differences between PAD5 5.976 11.22) showed increased PAD compared subgroups based on TBI severity, mechanism of injury, and the to controls (GM mean PAD5 0.076 7.41; WM mean presence of contusions. Analysis of PAD score was repeated for both GM and WM, and an exploratory analysis of the relation- PAD5 2.066 7.41; Fig 4). Comparing these PAD ship between the 2 brain tissue classes was also carried out. scores between TBI and control groups showed signifi- cant differences for both GM (F5 14.8, p< 0.001) and Results WM (F5 9.7, p5 0.002). The discrepancy between predicted and chronologi- Chronological Age Can Be Predicted from cal age was related to the severity of injury (Fig 5A, B). Structural Neuroimaging The model was able to accurately predict chronological Patients with a moderate/severe classification showed age for both training and control test data sets. For the increased mean PAD score (GM mean5 5.726 10.97; training set (Fig 3), age was accurately predicted based WM mean5 7.246 11.64), whereas mild (probable) on both GM (predicted–chronological age correlation patients were not significantly different from controls r5 0.921, R 5 0.848, MAE5 6.2 years) and WM (GM mean520.426 8.48; WM mean520.146 6.1). (r5 0.922, R 5 0.851, MAE5 6.16 years). Permutation The presence of focal lesions had no influence on PAD April 2015 575 ANNALS of Neurology FIGURE 4: Traumatic brain injury (TBI) patients show increased predicted age difference (PAD) score for gray matter and white matter. Boxplots of PAD score are shown, calculated by subtracting chronological age from predicted age for the test data sets of 99 TBI patients and 113 healthy controls. (A) PAD scores derived from the gray matter model showing a significant increase in TBI patients. (B) PAD scores from white matter also show an increase in TBI patients. score, as there were no significant differences between an increasing discrepancy between predicted and chrono- those with and without contusions for either GM (mean: logical age with a longer TSI (Fig 6). As there was also a 5.99 vs 5.55, p5 0.85) or WM (mean: 6.89 vs. 7.44, correlation between TSI and age (r5 0.28, p5 0.0047) p5 0.83). Furthermore, when limiting the TBI patient and between PAD score and age (GM: r5 0.18, group to only moderate/severe lesion-free patients p5 0.068; WM: r5 0.21, p5 0.037), a partial correla- (n5 31), ANCOVA for GM and WM still showed sig- tion approach was used to examine directly the relation- nificantly greater PAD scores compared to controls ship between PAD score and time since TBI, to remove (p< 0.001). Although mean PAD score for TBI patients variance associated with age. Three outliers in terms of and controls was greater for WM than GM, the relatively TSI were present in the TBI group (168, 206, and 563 high variability meant that this difference was not signifi- months), defined as being 62 SD away from the mean cant (p> 0.1). (mean5 28.386 63.58, 2-SD range5 0–155.54 months). These patients were not driving the association, Brain Atrophy Correlates with Cognitive as the results were still highly significant after these Impairment patients were removed (GM: r5 0.506, p< 0.001; WM: Performance on a number of cognitive measures corre- r5 0.467, p< 0.001). Removing patients with focal lated strongly with PAD scores (Table 2). After correc- lesions also did not alter the association between TSI and tion for the multiple tests conducted (using the false PAD score, as analysis in contusion-free patients (n5 47) discovery rate ), measures relating to information proc- was still significant (GM: r5 0.353, p5 0.01; WM: essing speed and memory were significantly correlated r5 0.342, p5 0.016). with GM and WM PAD, with increased reaction time being related to increases in PAD score. Executive func- Brain "Ageing" Does Not Vary in Patients with tion measures were significantly correlated for GM PAD Different Mechanism of Injury scores, but not for WM. Scores on the WASI subscales We assessed whether the mechanism of injury affected were not correlated with PAD scores. brain ageing. Patients who suffered TBI due to RTA, assault, or fall all showed increased mean PAD scores Evidence for Accelerated Atrophy following TBI There was a strong correlation between PAD score and (5.93, 6.15, and 4.11 for GM; 7.52, 4.42, and 6.51 for TSI for both GM (r5 0.535, p< 0.001) and WM WM), indicating that TBI patients’ brains appeared to (r5 0.496, p< 0.001), controlling for age. This reflected be older, irrespective of the cause of injury (see Fig 5C, 576 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI FIGURE 5: Gray matter (GM) and white matter (WM) predicted age difference (PAD) score, stratified by injury severity and injury mechanism. Boxplots of PAD score in the traumatic brain injury (TBI) patient group are shown, stratified by clinical char- acteristics. (A) GM PAD score distributions for each Mayo classification: probable/mild, moderate/severe, indicating that brain age is only increased in moderate/severe patients, not in mild TBI. (B) Mayo classification for WM. (C) GM PAD score by mech- anism of injury (assault, fall, road–traffic accident [RTA]), indicating that similar levels of increased brain ageing occur independ- ent of mechanism of injury. (D) Mechanism of injury for WM. D). There were no significant differences in PAD on average >4 years older than the patient’s chronologi- between mechanism subgroups for either GM or WM. cal age. This discrepancy was only seen in patients with The correlation between TSI and PAD was still present more severe injuries, was independent of the mechanism for patients injured during RTAs and assaults, although of injury, and was predictive of cognitive impairment. not for sufferers of falls. Mean TSI associated with falls These results support the theory proposed by Moretti was only 19.8 (635.2) months, considerably less than and colleagues that TBI may hasten the ageing process for RTA (33.26 43.1 months) and assaults and is in keeping with long-term structural and func- (41.56 120.7 months). This decreased duration and nar- tional brain abnormalities reported in neuroimaging and 11–13,25,26 rower distribution may have led to lower sensitivity to neuropathology studies of TBI patients. the effects of TSI on PAD in sufferers of falls. Animal work demonstrates a progressive loss of brain tissue after experimental TBI. A number of mech- Discussion anisms for this have been proposed, including chronic 12,26 Using a multivariate method to investigate the spatial neuroinflammation, Wallerian degeneration of patterns of age-associated brain atrophy, we show that WM, and the deposition of abnormal tau and TBI produces a pattern of structural brain changes that amyloid-b proteins. These processes lead to alterations affects the apparent brain age of both GM and WM. in cellular morphology, loss of trophic support, and even- The predictive model we generated was highly accurate tual cell death and gross atrophy. It is also possible that at estimating chronological age in healthy participants, a reduction in brain volume might be due to the clear- based only on the appearance of T1 MRI scans. In con- ance of tissue damaged at the time of injury. Previous trast, following TBI, the model estimated brain age to be neuroimaging studies have typically shown atrophy when April 2015 577 ANNALS of Neurology TABLE 2. Relationship between Neuropsychological Measures and PAD in TBI Patients Cognitive Neuropsychological Test No. Mean GM PAD, p WM PAD, p Domain rho rho a a Processing Trail Making Test A, s 90 29.59 (12.15) 0.338 0.001 0.379 <0.001 speed a a Trail Making Test B, s 90 68.79 (39.79) 0.343 0.001 0.271 0.009 a a Stroop color naming, s 90 34.47 (8.83) 0.279 0.007 0.281 0.006 a a Stroop word reading, s 90 23.93 (5.49) 0.269 0.009 0.306 0.003 Choice reaction task 66 478 (124) 0.331 0.005 0.243 0.047 median reaction time, ms Executive Trail Making 90 39.15 (32.51) 0.262 0.011 0.154 0.146 function Test B minus A, s Inhibition/switching, s 89 69.54 (22.18) 0.247 0.018 0.205 0.053 Inhibition/ 89 37.24 (18.41) 0.213 0.043 0.168 0.115 switching minus baseline Stroop performance, s Letter fluency total 89 39.01 (12.73) 20.271 0.009 20.133 0.214 Intellectual WASI similarities 90 37.04 (5.22) 20.160 0.132 20.136 0.200 ability WASI matrix reasoning 88 26.34 (5.53) 20.104 0.336 0.013 0.908 a a Memory People Test 90 22.94 (7.33) 20.253 0.015 20.254 0.014 immediate recall Correlations with PAD score were conducted with variance accounting for chronological age partialed out, using the Spearman rank-order approach. Denotes statistical significance after false discovery rate correction for multiple comparisons. GM5 gray matter; PAD5 predicted age difference score; TBI5 traumatic brain injury; WASI5 Wechsler Abbreviated Scale for Intelligence; WM5 white matter. comparing acute and chronic scans, making it difficult to age. Discrepancies between brain and chronological age disentangle these possibilities. One longitudinal study appear to be biologically informative, perhaps reflecting has shown that GM volume decreases continue between important differences in susceptibility or resistance to 1 and 4 years after TBI, demonstrating that tissue loss age-related pathology. This possibility is supported by continues well beyond the stage when clearance of the relationship we observed between PAD score and acutely damaged tissue is likely to have finished. This cognitive function. Our structural imaging measures cor- suggests a progressive process, which is supported by our related with cognitive impairment in domains typically 10,11 observation that patients who were assessed further from affected by TBI, namely information processing their injury (high TSI) showed more atrophy. This sup- speed, memory, and executive function. As the vast ports the idea that tissue loss accelerates over time, in majority of these patients were at least 3 months postin- keeping with a progressive neurodegenerative process jury, it is likely that such impairments are persistent, triggered by the injury (see Figs 1 and 4 and reviews by rather than acute and transient, implying that there is a 2 9 Smith and colleagues and Bigler ). relationship between the observed brain changes and the Our model explained much of the variation in chronic post-TBI cognitive profile. Intriguingly, these are chronological age, in line with previous research showing the same cognitive domains often affected by age-related that multivariate brain imaging analysis can accurately cognitive decline, which leads us to speculate that PAD 15,30 predict chronological age. Much like outwardly visi- score may be relevant to the effects of ageing in a ble signs of age such as the presence of wrinkles or gray broader sense. In particular, the strongest associations hair, this shows that the brain also varies in its apparent with our measure of brain ageing were found with 578 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI FIGURE 6: Gray matter and white matter predicted age difference (PAD) score increases with greater time since injury (TSI). Scatterplots depicting the relationship between PAD score (x-axis) and TSI (y-axis) are shown. Plotted values are the residuals derived from a linear regression with PAD score or TSI, regressing out chronological age. (A) Gray matter (dark gray circles) PAD scores, with dashed lines representing the locally weighted scatterplot smoothing (lowess) line calculated (dashed gray line). B) White matter (light gray triangles) PAD scores with lowess line (dashed light gray line). Both analy- ses were conducted after the removal of 3 outliers, identified based on having a TSI of 62 standard deviations from the mean. information processing speed measures, and impairments phy is important, as it suggests that the underlying in this domain have been posited as central to the cogni- differences in brain structure are unlikely to be secondary tive decline associated with typical ageing. The nature to nonspecific neurological or demographic factors that of our sample did not permit an exhaustive examination might be expected to vary across individuals with TBI of TBI-related and age-related cognitive dysfunction; from very different causes. For example, a possible con- however, our interpretation of results supports the idea found is that patients predisposed to TBI might have that the pattern of brain changes detected in TBI smaller brains prior to injury. However, there is no evi- patients has similar functional consequences to normal dence from other studies that this is the case, and it is ageing, albeit occurring in an accelerated form. Measures highly unlikely that patients with such contrasting mech- of age discrepancy might be useful for screening clinical anisms of injury would also show the same systematic and population samples to identify those at increased bias in brain size. risk of age-associated pathologies, for quantifying the It is unclear whether the brain atrophy we observe effects of vascular risk factors or neuropsychiatric diseases after TBI reflects ongoing neurodegeneration triggered by on general brain health, and for stratifying patients for the injury or an interaction with normal ageing. In part, targeted treatments or clinical trial enrollment. this reflects the complexity of defining ageing. Rates of As predicted, the degree of apparent ageing reflects ageing vary between individuals, but also can affect dif- the severity of initial injury. In contrast to moderate/ ferent tissues in the same person at different rates. This severe TBI, patients with minor TBI showed no brain makes comprehensively modeling age challenging. One atrophy. This suggests that a significant biomechanical limit of the work is that we focus only on a single facet force is necessary to trigger ongoing neurodegenerative of ageing, demonstrating that individuals have more processes that lead to progressive atrophy. In addition, brain atrophy after TBI. We are unable to disentangle the mechanism of injury did not influence PAD score, the degree to which this atrophy results from an accelera- with those suffering falls, assaults, and RTAs showing tion of processes seen in healthy ageing, whether the similar levels of atrophy. That widely varying mecha- injury triggers new neurodegenerative processes, or the nisms of TBI result in similar patterns of premature atro- extent to which these two possibilities are inter-related. April 2015 579 ANNALS of Neurology Processes involved in neurodegeneration such as inflam- that resemble the atrophy seen during ageing. The accel- mation and the accumulation of misfolded proteins are erated trajectory of brain atrophy we observed is in keep- also seen to varying degrees in healthy ageing, and com- ing with a progressive neurodegenerative process parative studies focusing on these factors will be neces- triggered by the injury. The effects of age and injury are sary to define to what extent TBI patients resemble likely to act synergistically, leading to deficits in informa- healthy older individuals or should be viewed as having tion processing speed and other neuropsychological meas- distinct neuropathology. ures known to be impaired in normal ageing. Future A number of other potential limitations of the studies could use PAD score as a predictor of clinical study should also be noted. First, some TBI patients outcome or as a surrogate marker of treatment efficacy. have small focal lesions that could potentially cause prob- The work gives insight into the etiology of the long-term lems with the image registration techniques that our age chronic effects of TBI and has implications for the long- prediction requires. However, we use advanced registra- term care and potential future treatments for TBI tion algorithms that perform well in our patient group, patients, as methods that attenuate the negative effects of and our results are similar when the analysis is restricted ageing may also be effective treatments for patients who to patients without focal lesions. Second, the estimates of have suffered a TBI. age were variable, indicating a degree of residual mea- surement error, which is to be expected as we were using Acknowledgment only brain structure to predict chronological age. Never- This research was supported by a European Union Sev- theless, the large sample size meant we were sufficiently enth Framework Programme grant to the Comorbidity powered to detect statistically significant group differen- in Relation to AIDS (COBRA) project (FP-7-HEALTH ces in PAD score. 305522; J.H.C.), an National Institute for Health We exploited the wide range of TSI to explore the Research (NIHR) Professorship (NIHR-RP-011-048; chronic effects of TBI. Our analysis indicated both that D.J.S.), and the NIHR Imperial Biomedical Research increased atrophy is present after TBI, and also that the Centre. amount of atrophy increases over time. Removing the Data used in the preparation of this article were patients with the greatest TSI did not alter the results, obtained from the Alzheimer’s Disease Neuroimaging implying that this apparent acceleration is a robust effect Initiative (ADNI) database (http://adni.loni.usc.edu/). As even over a relatively short time period. As there are a such, the investigators within ADNI contributed to the number of potential confounds inherent with cross- design and implementation of ADNI and/or provided sectional analyses, our evidence for accelerating atrophy data but did not participate in analysis or writing of this should be interpreted with caution. For example, brain report. The draft manuscript was certified by the ADNI so ageing is likely to be in part genetically mediated, Data and Publications Committee as meeting the ADNI information about genotype should be added in future data sharing requirements. Full details of the ADNI data studies to potentially explain even greater amounts of sharing policy and further information fully acknowledg- variance in age. In addition, the severity of the initial ing the relevant funding sources can be found here: injury will influence the extent of brain atrophy. We only http://adni.loni.usc.edu/wp-content/uploads/how_to_ observed increased atrophy in patients with moderate/ apply/ADNI_DSP_Policy.pdf severe injuries, which shows the expected impact of The views expressed are those of the authors and not injury on apparent brain age. However, within the mod- necessarily those of the National Health Service, the erate/severe group there will be variations in injury sever- NIHR, or the Department of Health. ity, and these could potentially confound the relationship between TSI and brain atrophy. Individual indices of injury severity such as Glasgow Coma Scale score and duration of post-traumatic amnesia are known to be Potential Conflicts of Interest crude measures of injury severity. Therefore, large lon- D.J.S.: investigator led grant, Pfizer. gitudinal studies will be necessary to completely resolve these issues, ideally with individual follow-up at multiple time points to allow longitudinal within-subject analysis that incorporates other genetic and biomarker informa- References tion about the ageing process. 1. Smith DH, Chen XH, Pierce JES, et al. Progressive atrophy and In summary, our study is the first empirical in vivo neuron death for one year following brain trauma in the rat. demonstration that TBI causes structural brain changes J Neurotrauma 1997;14:715–727. 580 Volume 77, No. 4 Cole et al: Brain Age Prediction after TBI 2. Smith DH, Johnson VE, Stewart W. Chronic neuropathologies of 19. Koutsouleris N, Davatzikos C, Borgwardt S, et al. Accelerated single and repetitive TBI: Substrates of dementia? Nat Rev Neurol brain aging in schizophrenia and beyond: a neuroanatomical 2013;9:211–221. marker of psychiatric disorders. Schizophr Bull 2014;40:1140– 3. Moretti L, Cristofori I, Weaver SM, et al. Cognitive decline in older adults with a history of traumatic brain injury. Lancet Neurol 2012; 20. Salthouse TA. The processing-speed theory of adult age differen- 11:1103–1112. ces in cognition. Psychol Rev 1996;103:403–428. 4. Bazarian JJ, Cernak I, Noble-Haeusslein L, et al. Long-term neuro- 21. Ashburner J. A fast diffeomorphic image registration algorithm. logic outcomes after traumatic brain injury. J Head Trauma Reha- Neuroimage 2007;38:95–113. bil 2009;24:439–451. 22. Schrouff J, Rosa MJ, Rondina JM, et al. PRoNTo: pattern recogni- 5. Hedden T, Gabrieli JDE. Insights into the ageing mind: a view tion for neuroimaging toolbox. Neuroinformatics 2013;11:319– from cognitive neuroscience. Nat Rev Neurosci 2004;5:87–96. 337. 6. Ferreira D, Molina Y, Machado A, et al. Cognitive decline is medi- 23. Doyle OM, Ashburner J, Zelaya FO, et al. Multivariate decoding ated by gray matter changes during middle age. Neurobiol Aging of brain images using ordinal regression. Neuroimage 2013;81: 2014;35:1086–1094. 347–357. 7. Sowell ER, Peterson BS, Thompson PM, et al. Mapping cortical 24. Benjamini Y, Hochberg Y. Controlling the false discovery rate—a change across the human life span. Nat Neurosci 2003;6:309–315. practical and powerful approach to multiple testing. J R Stat Soc B Met 1995;57:289–300. 8. Raz N, Rodrigue KM. Differential aging of the brain: patterns, cog- nitive correlates and modifiers. Neurosci Biobehav Rev 2006;30: 25. Povlishock JT, Katz DI. Update of neuropathology and neurologi- 730–748. cal recovery after traumatic brain injury. J Head Trauma Rehabil 2005;20:76–94. 9. Bigler ED. Traumatic brain injury, neuroimaging, and neurodegen- eration. Front Hum Neurosci 2013;7:395. 26. Johnson VE, Stewart JE, Begbie FD, et al. Inflammation and white matter degeneration persist for years after a single traumatic brain 10. Draper K, Ponsford J. Cognitive functioning ten years following injury. Brain 2013;136:28–42. traumatic brain injury and rehabilitation. Neuropsychology 2008; 22:618–625. 27. Meythaler JM, Peduzzi JD, Eleftheriou E, Novack TA. Current con- cepts: diffuse axonal injury-associated traumatic brain injury. Arch 11. Kinnunen KM, Greenwood R, Powell JH, et al. White matter dam- Phys Med Rehabil 2001;82:1461–1471. age and cognitive impairment after traumatic brain injury. Brain 2011;134:449–463. 28. Johnson VE, Stewart W, Smith DH. Widespread tau and amyloid- beta pathology many years after a single traumatic brain injury in 12. Ramlackhansingh AF, Brooks DJ, Greenwood RJ, et al. Inflamma- humans. Brain Pathol 2012;22:142–149. tion after trauma: microglial activation and traumatic brain injury. Ann Neurol 2011;70:374–383. 29. Farbota KD, Sodhi A, Bendlin BB, et al. Longitudinal volumetric changes following traumatic brain injury: a tensor-based mor- 13. Sharp DJ, Beckmann CF, Greenwood R, et al. Default mode net- phometry study. J Int Neuropsychol Soc 2012;18:1006–1018. work functional and structural connectivity after traumatic brain injury. Brain 2011;134:2233–2247. 30. Groves AR, Smith SM, Fjell AM, et al. Benefits of multi-modal fusion analysis on a large-scale dataset: life-span patterns of inter- 14. Ross DE. Review of longitudinal studies of MRI brain volumetry in subject variability in cortical morphometry and white matter micro- patients with traumatic brain injury. Brain Inj 2011;25:1271–1278. structure. Neuroimage 2012;63:365–380. 15. Franke K, Ziegler G, Kloppel € S, Gaser C. Estimating the age of 31. Simm A, Johnson TE. Biomarkers of ageing: a challenge for the healthy subjects from T1-weighted MRI scans using kernel meth- future. Exp Gerontol 2010;45:731–732. ods: exploring the influence of various parameters. Neuroimage 2010;50:883–892. 32. Filippini N, Ebmeier KP, MacIntosh BJ, et al. Differential effects of the APOE genotype on brain function across the lifespan. Neuro- 16. Brown T, Kuperman JM, Chung Y, et al. Neuroanatomical assess- image 2011;54:602–610. ment of biological maturity. Curr Biol 2012;22:1693–1698. 33. Malec JF, Brown AW, Leibson CL, et al. The mayo classification 17. Dosenbach NUF, Nardos B, Cohen AL, et al. Prediction of individ- system for traumatic brain injury severity. J Neurotrauma 2007;24: ual brain maturity using fMRI. Science 2010;329:1358–1361. 1417–1424. 18. Gaser C, Franke K, Kloppel € S, et al. BrainAGE in mild cognitive impaired patients: predicting the conversion to Alzheimer’s dis- ease. PLoS One 2013;8:e67346. April 2015 581

Journal

Annals of Neurology – Pubmed Central

Published: Mar 25, 2015

There are no references for this article.