If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Acquired Brain Injury (ABI) describes a range of brain injuries occurring after birth, including tumor, traumatic brain injury or stroke. Although MRIs are routinely used for diagnosis, prediction of outcome following brain injury is challenging. Quantitative structural information from brain images may provide an opportunity to predict patient outcomes; however, due to the high prevalence of severe pathology in children with ABI, quantitative approaches must be robust to injury severity.
In this pilot cross-sectional study, automated quantitative measures were extracted from the MRIs of a cohort of children with ABI (n = 30, 8–16 years, follow up MRI taken 1.8–13.4 years after time of injury) as well as 36 typically developing controls with no brain injury (7–17 years) using a pathology-robust technique. Measures of brain volume, lesion volume and cortical morphology were associated with concurrent motor, behavioral, visual and communicative function using Least Absolute Shrinkage and Selection Operator (LASSO) regression.
These regression models were validated on a separate test set (n = 8 of the ABI cohort), which revealed significant correlations between measures of brain structure with motor, cognitive, visual and communicative function (r = 0.65–0.85, all p < 0.01). Furthermore, comparisons of the structural measures to the typically developing cohort revealed overall reductions in global grey matter volume among the ABI cohort, as well as cortical thinning in several cortical areas.
These preliminary associations reveal that motor and behavioral function can be estimated from MRI alone, highlighting the potential utility of the proposed pathology-robust MRI quantification tools to provide estimates of long-term clinical prognosis of children with ABI following injury.
Acquired brain injury (ABI) is an umbrella term encompassing traumatic brain injury and cerebral vascular incidents after 28 days post full-term birth, and refers to non-progressive and non-degenerative brain injury affecting neurological function.
Brain measurements have revealed significantly reduced brain volumes and significantly increased volumes of cerebrospinal fluid (CSF) in patients with ABI compared to typically developing children (TDC),
Quantifying brain structure and lesions from MRI has the potential to assist in the clinical diagnosis of ABI, as well as predicting patient outcomes, which is critical for tailoring a patient-specific treatment strategy.
Despite this, the automated quantification of brain structure from MRI remains challenging due to the severity of brain injury that may be present in patients with ABI. Furthermore, secondary effects such as hydrocephalus may be severe, and may differ so significantly from healthy anatomy that any a priori information may not be applicable and may negatively impact the performance of algorithmic approaches.
However, patients with severe brain lesions potentially have the greatest functional impairments and would benefit most from a clinical prognosis leading to tailored intervention. Automated methods that are robust to brain injury are needed to include such patients for quantitative analysis.
In this pilot study, we aim to quantify brain structure on a cohort of children with ABI (n = 30) using a pathology-robust automated segmentation pipeline previously validated in a cohort of children with cerebral palsy.
Using this pipeline, we aim to identify the structural differences between children with ABI and a cohort of typically developing children (TDC) (n = 36), to characterize the structural differences caused by the heterogeneous injury observed in ABI. Furthermore, we aim to extract the associations between brain structure and motor, cognitive, visual and communicative function on the preliminary ABI cohort using regression models, so that predictive MRI-derived biomarkers of outcome may be identified.
Imaging and clinical data of a cohort of children with ABI were acquired by the Queensland Cerebral Palsy and Rehabilitation Research Centre (QCPRRC)
between 2013 and 2014 (Australian clinical trials register ACTRN12613000403730). Children were included if they were between 8 and 16 years, were at least 12 months post ABI diagnosis, were Gross Motor Classification System (GMFCS) I or II (i.e., were independently ambulant) and were classified with a brain injury greater than 28 days after birth. Children were excluded from the study if they had degenerative or metabolic conditions, unstable epilepsy (not controlled by medication) or they had undergone surgical or medical interventions impacting their upper and lower limb function. A total of n = 60 children were recruited as part of the Mitii-ABI study. Participants were scanned 1.8–13.3 years after ABI diagnosis, with all receiving standard care prior to recruitment in the study. Imaging and clinical data from the TDC cohort came from an associated study with identical imaging protocols
recruited between 2011 and 2014 (Australian clinical trials register ACTRN12611001174976).
2.2 Image acquisition
A subset of ABI participants (n = 30) underwent T1 Magnetization Prepared Rapid Gradient Echo (MPRAGE) scanning on a 3T Siemens MAGNETOM Trio Tim scanner with scanning parameters (TR = 1900 ms, TE = 2.32 ms, TI = 900 ms, flip angle = 9°, image resolution = 0.89 x 0.89 × 0.9 mm). Participants also underwent an axial T2 Turbo Inversion Recovery Magnitude (TIRM) (TR = 7000 ms, TE = 79 ms, flip angle = 120°, image resolution = 0.43 x 0.43 × 5.2 mm).
2.3 Image processing
MPRAGE and TIRM sequences underwent pre-processing steps, including N4 bias correction, image smoothing using anisotropic diffusion and affine alignment to the Colin 27 atlas using ANTS registration. Images were checked for motion artefact, with n = 2 found to contain moderate artefact, n = 5 with mild artefact, and n = 23 with no visible artefact.
Tissue segmentation into grey matter (GM), white matter (WM) and CSF was then performed using Expectation Maximization – Markov Random Fields (EM-MRF)
on the MPRAGE. Unlike atlas-driven approaches, this approach is robust to severe tissue loss or tumor resection. Using the TIRM sequence, an additional Expectation Maximisation (EM) segmentation was performed to obtain a lesion segmentation.
Using the obtained CSF segmentation, the lateral ventricles can be identified using flood filling from atlas-derived seed locations of the ventricle. Measures including cortical thickness, sulcal depth and cortical curvature can be quantified from the cortical GM segmentation, in each cortical region defined in the Automated Anatomical Labeling (AAL) atlas.
In addition, the regional deep grey matter (DGM), including the caudate nucleus, globus pallidus and thalamus, and the corpus callosum (CC), using a novel patch-based approach.
The final segmentations were also manually checked for anatomical accuracy. This entire processing pipeline is illustrated in Fig. 1.
2.4 Clinical assessments
Of the Mitii-ABI assessment battery, a subset of clinical scores with a large dynamic range and normal distribution were chosen to capture their sensorimotor, visual perceptual, executive functioning and verbal reasoning functioning. As a measure of upper-limb motor function, the Assisting Hand Assessment (AHA) was recorded, which reliably measures the performance of the assisting (i.e., impaired) hand in bimanual tasks.
was used to assess participants’ fluid reasoning and verbal comprehension. Clinical assessments were performed prior to MR imaging, either on the same day or day before scanning. TDC did not undergo any clinical assessments.
2.5 Statistical methodology
To elucidate the structural differences caused by the ABI, each quantified structural measure was compared between ABI and TDC cohorts using Student's t-tests, correcting for multiple comparisons with Bonferroni correction. To determine any difference in age between cohorts, Student's t-test was performed, while the Chi-squared test was used to determine any sex balance differences between cohorts. All MRI measures were checked for normality using the Kolmogorov–Smirnov (K-S) test for normality prior to group comparisons.
To investigate associations between brain structure and clinical function in the ABI cohort only, the Least Absolute Shrinkage and Selection Operator (LASSO) method was used with MRI measures being the independent variables and clinical function being the dependent variable. Patient age, biological sex and handedness, time since injury, MRI motion artefact strength (good, fair, poor, unusable) and the diagnosis of injury were included as covariates in all LASSO models. These categories include cardiovascular accident (CVA), TBI, tumor, encephalitis, and empyema. To minimize model overfitting due to the large number of structural measures, LASSO implicitly performs variable selection; hence structural biomarkers not strongly associated with the outcomes or that were collinear were automatically removed from the model, leaving measures that are independently predictive of the outcome. For all models, data was partitioned randomly into two age- and sex-matched groups, with models trained on 75% of the data (n = 22) and their performance validated on the independent 25% test set (n = 8). Test set correlations were computed using Pearson's r correlation for the LASSO models and were corrected for multiple comparisons using Bonferroni correction (adjusted alpha of 0.0125, for 4 regression model comparisons).
The demographics of the children with ABI and the TDC in this cohort are provided in Table 1. Thirty children with ABI were included in the study (age range 8–16 years, 20 male). Aetiologies included cerebrovascular accident (n = 13, 43%), traumatic brain injury (n = 10, 33%) and tumor (n = 4, 13%). The TD cohort consisted of 36 typically developing children (age range 7–17 years, 19 male), who were both age- and gender-matched to the ABI cohort. Using the Shapiro–Wilk test of normality, time since injury was found to be normally distributed (p = 0.34).
including ABI patients with severe injury, as shown in Fig. 2. The observed structural differences between the ABI and TDC cohorts are detailed in Supplementary Table 1, revealing reduced global GM volume and corpus callosum splenium volume. Cortical thickness was also observed to be reduced in several regions (including the supplementary motor area, precuneus, paracentral lobule and the middle temporal gyrus) as well as increased sulcal depth observed in the precentral and postcentral gyri, reflecting potential cortical GM atrophy. The differences in several of these structure measures are illustrated in Supplementary Fig. 1.
3.3 Observed structure-function relationships
The image measures retained from LASSO, with their respective regression coefficients for each of the four models, are given in Table 2. In most of the LASSO models, patient covariates including patient age, biological sex, ABI aetiology, handedness, motion artefact severity in the T1 MRI, and time since injury were dropped, indicating that these covariates were not as strongly associated with any of the clinical outcomes as brain morphology.
Table 2The retained anatomical regions, and corresponding regression coefficients (including standard errors) of the four LASSO regression models modelled on the 75% training set. For each model, the multiple R-squared is provided.
Caudate nucleus volume
CT of rolandic operculum
SD of superior frontal gyrus
SD of middle frontal gyrus
Curvature of precentral gyrus
Curvature of supramarginal gyrus
Cerebellar grey matter volume
CT of inferior frontal gyrus
CT of superior frontal gyrus
CT of paracentral lobule
SD of insula
SD of superior temporal gyrus
Curvature of inferior frontal gyrus
Curvature of middle frontal gyrus
Curvature of superior frontal gyrus
Curvature of cingulate
Curvature of angular gyrus
Amygdala lesion volume
CT of posterior cingulate gyrus
CT of angular gyrus
SD of medial superior frontal gyrus
SD of heschl gyrus
Curvature of inferior parietal gyrus
CT of inferior frontal gyrus
CT of insula
CT of calcarine fissure
CT of cuneus
CT of superior parietal gyrus
CT of paracentral lobule
SD of superior frontal gyrus
AHA, Assisting Hand Assessment; ALIC, anterior limb of the internal capsule; BRIEF, Behaviour Rating Inventory of Executive Function; CT, cortical thickness; SD, sulcal depth; TVPS, Test of Visual Perception Skills; WR, Word Reasoning.
The performance of the predictive regression models are shown in Table 3. All four of the data-driven models were statistically significant (p < 0.0125) in the test set. The correlation between the LASSO model predictions with the actual outcomes on the 25% test set are provided in Supplementary Fig. 2.
Table 3Associations between the LASSO models predicted and measured clinical outcomes on the test dataset using Pearson's r correlation and root mean square error.
Normalised root mean square error (%)
Asterisked model correlations were found to be statistically significant after correction for multiple comparisons: ∗p < 0.0125.
AHA, Assisting Hand Assessment; BRIEF, Behaviour Rating Inventory of Executive Function; TVPS, Test of Visual Perception Skills; WR, Word Reasoning.
In this study, pathology-robust approaches were applied to a cohort of children with ABI who were compared to a cohort of typically developing children. A direct comparison of structural measures revealed significantly reduced GM volume, corpus callosum splenium volume and reduced cortical thickness in several cortical regions, highlighting widespread atrophy in the ABI cohort, which is consistent with previous findings.
The more frequent prevalence of cortical thickness differences when compared to controls (Supplementary Table 1) and in association with outcomes (Table 2) indicate that neuronal density impacted by tissue loss is more informative on this cohort. Although the ABI and TD cohorts in this study were both age- and sex-matched, there is little evidence observing different incidence of ABI between sexes,
and indeed sex was not found to be significantly associated with poorer clinical function in the present study, being dropped in all LASSO models in Table 2.
The LASSO models can provide an insight into the observed relationships between brain structure and function within the ABI cohort. The structural measures retained by LASSO are provided in Table 2, with several of these regions having known roles in the clinical function that they were associated with, such as the thalamus, the precentral gyrus,
all important areas involved in the motor pathway, being retained in the AHA model. For the BRIEF that assesses behavioral manifestations of executive function in everyday life, retained regions in the model include the insula which has a role in emotion and cognitive functioning,
In addition, the superior, middle and inferior frontal gyri were retained, which collectively with the insula, cingulate and amygdala, form the basis of the salience and emotion network which regulates emotional salience.
The WR subtest assesses inductive reasoning using verbal stimuli, which may explain why a large set of brain regions were retained that are associated with inductive reasoning, including the inferior and superior frontal gyrus,
Even the TVPS model which retained the fewest predictive biomarkers, identified regions involved in the ventral visual pathway that the TVPS measure assesses, including the inferior parietal gyrus which controls visuotemporal attention
As a subset of the retained features are consistent with prior knowledge of the biology, these models may be elucidating true structure-function relationships in the brain. This is further supported by the significant test set correlations, indicating that the models can estimate these clinical scores from unseen data and may indeed generalize to new data.
The main limitation of this study is the modest size of the ABI dataset, particularly given the heterogeneity of ABI aetiology and the time since injury when participants were recruited into the Mitii-ABI study. Some aetiologies of ABI are very rare in this cohort (Table 1), with conditions such as encephalitis (n = 2 in this study) having different aetiologies, and tumors (n = 4) being either benign or malignant, which all may affect the sMRI characteristics. As a result, the associations between structural brain measures and clinical function need to be interpreted with care and require validation on a larger cohort, with a focus on TBI and CVA categories, to demonstrate generalizability to the wider ABI population. Additionally, this study only utilizes the structural T1-and T2-weighted MRIs; however, given the importance on neuroinflammation phenomena on recovery and patient outcomes,
will be key for prediction of outcomes from MRI. Finally, the cross-sectional design of this study with variable times since injury means that only the prediction of concurrent patient function can be estimated from quantitative MRI, which is not at the level of translational clinical relevance. Future study performing longitudinal MRI acquisitions at fixed time points can be used to determine which stages of injury (acute, subacute or chronic) would be more valuable in the prediction of long-term outcomes since early imaging. Ideally, early prediction of lesion definition over time and its impact on long-term function will be able to guide early tailored decisions around intervention strategies for patients with potential benefit on outcome. However, without such an ABI cohort, this work lays the methodological foundation for predicting future outcomes from MRI taken shortly after the time of injury, and future work will investigate the use of these techniques with earlier scans to guide decisions around intervention strategies for patients.
In this pilot study, automated quantification of structural MRIs has been performed on a cohort of children with ABI of various aetiologies and with potentially severe brain injury. These measures of brain structure were compared to a typically developing cohort to characterize the extent of brain injury, which revealed significant reductions in global GM and corpus callosum volumes, as well as thinning in several cortical areas. Associations with clinical outcomes using LASSO revealed structure-function relationships consistent with the known physiology of the brain, despite use of feature selection based on data-driven approaches rather than a priori information. Importantly, these models were found to accurately estimate these clinical scores on an unseen partition of the data, indicating that motor, behavioral and communicative function could be predicted from new structural MR images. These preliminary findings highlight the utility of pathology-robust quantitative radiological assessment of children with ABI, and encourage future work to validate this approach on larger, longitudinal cohorts which can facilitate estimation of long-term prognosis, allowing for the targeting of optimal interventions early in life.
Alex M Pagnozzi: Methodology, Software, Validation, Formal analysis, Visualisation, Writing – Original Draft.
Kerstin Pannek: Methodology, Supervision, Writing – Reviewing and Editing.
Jurgen Fripp: Investigation, Project administration, Writing – Reviewing and Editing.
Simona Fiori: Methodology, Investigation, Writing – Reviewing and Editing.
Roslyn N. Boyd: Conceptualisation, Investigation, Project administration, Writing – Reviewing and Editing.
Stephen Rose: Conceptualisation, Methodology, Investigation, Writing – Reviewing and Editing.
For these studies involving human participants (ACTRN12611001174976 and ACTRN12613000403730 respectively), written and informed consent was obtained from the parent or legal guardian of each child. Furthermore, all procedures performed in these studies were in accordance with the ethical standards of the Medical Ethics Committee of The University of Queensland (2011000608, and 2013000212), and The Royal Children's Hospital Brisbane (HREC/11/QRCH/35, and HREC/12/QRCH/222), and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
Consent to participate
Written and informed consent was obtained from all participants and their parents or guardians to participate in the study.
Consent to publish
Consent to publish group-level findings was obtained from all participants and their parents or guardians.
Alex M. Pagnozzi is supported by the Advance Queensland Research Fellowship (AQR16816-17RD2). This funding body has not contributed to the study design, the collection, management, analysis and interpretation of data, the writing of final reports or the decision to submit findings for publication. We thank Dr Olga Laporta-Hoyos for her feedback on the article.
Availability of data and materials
According to ethics of the Mitii-ABI (HREC/12/QRCH/222) and Mitii (HREC/11/QRCH/35) studies, the ethics form does not select these data to be made available to a third party. As a result of these ethics, making these data accessible to a third party will require new ethical approval to be sought, and additionally require consent from the families. As CIA of the MiTii and Mitii-ABI studies, Roslyn Boyd ( [email protected] ) would be the contact to whom any request for data access may be submitted.
Declaration of Competing Interest
The authors have no potential conflicts of interest to declare.
Alex M. Pagnozzi is supported by the Advance Queensland Research Fellowship ( AQR16816-17RD2 ). This funding body has not contributed to the study design, the collection, management, analysis and interpretation of data, the writing of final reports or the decision to submit findings for publication. We thank Dr Olga Laporta-Hoyos for her feedback on the article.