Periodic Reporting for period 3 - PROBIt (Identifying Predictors of Risk and Resilience for poor neuropsychological Outcome following childhood Brain InsulTs (PROBIt))
Berichtszeitraum: 2019-09-01 bis 2021-02-28
The impact of insults to the developing brain upon cognition and behaviour has far-reaching consequences for the child, their family, education and health care systems, and government expenditure. Many variables (illness, environmental) contribute to different outcomes following similar insults, and they exert their influence via the child’s developing brain. Predicting which child will recover from early brain insult and identifying those at risk of poor outcome represents a major challenge, with significant health economic implications. An unexplored question is whether direct measurement of the structure and function of the developing brain can improve our ability to predict outcomes in the long-term. Thus, PROBIt aimed to assess the utility of brain imaging biomarkers to predict individual neuropsychological and neurobehavioural outcomes following paediatric brain injury, and to identify those factors that combine optimally to classify outcomes. The proposal adopts an unorthodox approach of combining heterogeneous injury groups to explore the structural and functional consequences of perturbing developing brain networks. PROBIt sought to integrate data from clinically relevant paediatric cognitive and behavioural assessment, neuroimaging and computational modelling in large cohorts of children with brain insults. Multivariate pattern analysis was to be used to train a statistical classifier to reliably predict individual child outcomes across three core domains: achievement, behaviour and cognitive ability.
PROBIt's aims were addressed through a series of planned objectives including collection of brain imaging data, longutudinal follow up assessment of child neurodevelopmental status, and development of novel analytic methods. Due to early termination of the award the overarching goal could not be delivered. However, considerable planned work was completed and key findings include:
* We developed a new metric, Developmental Divergence, for structural brain scans that enables deviation from typical brain network development to be measured and then used to predict child neurobehavioural outcomes
* We identified that Deep Learning tools developed successfully in adult neuroimaging studies were not applicable to children, and identified ponential solutions to this
* We established a novel measure of network dysconnectivity that could be used in clinically acquired brain MRI scans to predict child neurobehavioural outcomes
* We showed that clinically acquired imaging data could be used to generate structural connectivity network maps in the absence of advanced imaging (i.e. research scans) that are not available in standard paediatric imaging centres thereby expanding the applicability of our methods such that they could be readily translated to the field.
* We identified that in the absence of group differences in white matter brain changes, novel imaging methods were more sensitive to microstructural change following early life traumatic brain injury, providing our broader cohort with a sensitive tool to measure brain development.
* We reported an advance in the field on structural covariance network analysis in which we showed a novel method that yields individual SCN maps, whereas work to date has only been possible at the group level.
PROBIt's aims were addressed through a series of planned objectives including collection of brain imaging data, longutudinal follow up assessment of child neurodevelopmental status, and development of novel analytic methods. Due to early termination of the award the overarching goal could not be delivered. However, considerable planned work was completed and key findings include:
* We developed a new metric, Developmental Divergence, for structural brain scans that enables deviation from typical brain network development to be measured and then used to predict child neurobehavioural outcomes
* We identified that Deep Learning tools developed successfully in adult neuroimaging studies were not applicable to children, and identified ponential solutions to this
* We established a novel measure of network dysconnectivity that could be used in clinically acquired brain MRI scans to predict child neurobehavioural outcomes
* We showed that clinically acquired imaging data could be used to generate structural connectivity network maps in the absence of advanced imaging (i.e. research scans) that are not available in standard paediatric imaging centres thereby expanding the applicability of our methods such that they could be readily translated to the field.
* We identified that in the absence of group differences in white matter brain changes, novel imaging methods were more sensitive to microstructural change following early life traumatic brain injury, providing our broader cohort with a sensitive tool to measure brain development.
* We reported an advance in the field on structural covariance network analysis in which we showed a novel method that yields individual SCN maps, whereas work to date has only been possible at the group level.
"In the first reporting period we have established an excellent team of young scholars. PROBIt attracts interest from a range of potential collaborators and although we anticipated recruiting from one site in the UK, we have the potential to open up at four others should the need arise.
Work on our existing dataset has led to new analysis methods being developed in the group. This dataset was always intended to act as a ‘test bed’ for analysis pipelines and it has proved extremely fruitful. The PROBIt team will submit eight manuscripts arising directly from our first 18 months work for publication this year. These are in addition to the planned deliverables set out in the original application.
We are conducting new qualitative research to establish current practice in our field so that we are in a position to measure the impact of the eventual study results on the intended stakeholders and society more generally.
In the second reporting period we have achieved the following scientific insights:
We identified that Deep Learning approaches, including convolution neural networks used successfully in adults, to segment brain lesions in the cohort perform poorly in children. We have also detected that a commonly used semi-automated pipeline to measure cortical thickness from structural MRI scans is affected by the presence of lesions in the cases processed, with errors introduced in the contralesional brain homologue, potentially undermining a large body of published literature. We have developed a new method to establish individual level structural covariance networks as well as reporting on standard SCN in paediatric TBI for the first time. Advanced analysis of white matter connectivity in this group shows that novel analytic methods provide additional information compared to standard analyses and our work combining lesion segmentation with diffusion metrics suggests that we can measure the role of perturbed connectivity in later outcomes (i.e. disconnectome symptom mapping). To establish our machine learning approach, we have examined solutions to class imbalance and a paper demonstrating the value of this in predicting paediatric TBI outcomes is in preparation. The first instantiation of ""Developmental Divergence Index” is under review, based on our work using structural MRI data. We also report novel findings about factors associated with specific behaviours in child TBI and have published a systematic review of morphometric MRI studies in paediatric TBI as a result of this work. We have comprehensively established systematic and replicable study operating procedures, databases, and image analysis pipelines. We have a clear plan to develop replicable analytic frameworks for our neuropsychological data, creating a solid scientific base upon which to complete the remaining scientific tasks.
The final reporting period was an early temination to the award and included a period affected by Covid delays to research activity. During this phase, we published new findings, completed methods development but we unable to continue outcome assessments as planned. As this was a longitudinal study, the final outcomes were not yet obtained at the point of grant termination."
Work on our existing dataset has led to new analysis methods being developed in the group. This dataset was always intended to act as a ‘test bed’ for analysis pipelines and it has proved extremely fruitful. The PROBIt team will submit eight manuscripts arising directly from our first 18 months work for publication this year. These are in addition to the planned deliverables set out in the original application.
We are conducting new qualitative research to establish current practice in our field so that we are in a position to measure the impact of the eventual study results on the intended stakeholders and society more generally.
In the second reporting period we have achieved the following scientific insights:
We identified that Deep Learning approaches, including convolution neural networks used successfully in adults, to segment brain lesions in the cohort perform poorly in children. We have also detected that a commonly used semi-automated pipeline to measure cortical thickness from structural MRI scans is affected by the presence of lesions in the cases processed, with errors introduced in the contralesional brain homologue, potentially undermining a large body of published literature. We have developed a new method to establish individual level structural covariance networks as well as reporting on standard SCN in paediatric TBI for the first time. Advanced analysis of white matter connectivity in this group shows that novel analytic methods provide additional information compared to standard analyses and our work combining lesion segmentation with diffusion metrics suggests that we can measure the role of perturbed connectivity in later outcomes (i.e. disconnectome symptom mapping). To establish our machine learning approach, we have examined solutions to class imbalance and a paper demonstrating the value of this in predicting paediatric TBI outcomes is in preparation. The first instantiation of ""Developmental Divergence Index” is under review, based on our work using structural MRI data. We also report novel findings about factors associated with specific behaviours in child TBI and have published a systematic review of morphometric MRI studies in paediatric TBI as a result of this work. We have comprehensively established systematic and replicable study operating procedures, databases, and image analysis pipelines. We have a clear plan to develop replicable analytic frameworks for our neuropsychological data, creating a solid scientific base upon which to complete the remaining scientific tasks.
The final reporting period was an early temination to the award and included a period affected by Covid delays to research activity. During this phase, we published new findings, completed methods development but we unable to continue outcome assessments as planned. As this was a longitudinal study, the final outcomes were not yet obtained at the point of grant termination."
Current practice in paediatric neuroimaging excludes or ignores the impact of brain lesions on quantitative measures of brain structure and function. This work is highly challenging and our team was developing new approaches to retain those data so that we could achieve our goal of predicting outcomes at the individual child level. We developed new methods for structural brain image analysis that will be applicable outside of this project.