Connectome mapping with edge density imaging differentiates pediatric mild traumatic brain injury from typically developing controls: proof of concept

Artificial Intelligence in Pediatric Emergency Medicine: Applications, Challenges, and Future Perspectives

The dawn of Artificial intelligence (AI) in healthcare stands as a milestone in medical innovation. Different medical fields are heavily involved, and pediatric emergency medicine is no exception. We conducted a narrative review structured in two parts. The first part explores the theoretical principles of AI, providing all the necessary background to feel confident with these new state-of-the-art tools. The second part presents an informative analysis of AI models in pediatric emergencies. We examined PubMed and Cochrane Library from inception up to April 2024. Key applications include triage optimization, predictive models for traumatic brain injury assessment, and computerized sepsis prediction systems. In each of these domains, AI models outperformed standard methods. The main barriers to a widespread adoption include technological challenges, but also ethical issues, age-related differences in data interpretation, and the paucity of comprehensive datasets in the pediatric context. Future feasible research directions should address the validation of models through prospective datasets with more numerous sample sizes of patients. Furthermore, our analysis shows that it is essential to tailor AI algorithms to specific medical needs. This requires a close partnership between clinicians and developers. Building a shared knowledge platform is therefore a key step.

MaPPeRTrac: A Massively Parallel, Portable, and Reproducible Tractography Pipeline

Large-scale diffusion MRI tractography remains a significant challenge. Users must orchestrate a complex sequence of instructions that requires many software packages with complex dependencies and high computational costs. We developed MaPPeRTrac, an edge-centric tractography pipeline that simplifies and accelerates this process in a wide range of high-performance computing (HPC) environments. It fully automates either probabilistic or deterministic tractography, starting from a subject's magnetic resonance imaging (MRI) data, including structural and diffusion MRI images, to the edge density image (EDI) of their structural connectomes. Dependencies are containerized with Singularity (now called Apptainer) and decoupled from code to enable rapid prototyping and modification. Data derivatives are organized with the Brain Imaging Data Structure (BIDS) to ensure that they are findable, accessible, interoperable, and reusable following FAIR principles. The pipeline takes full advantage of HPC resources using the Parsl parallel programming framework, resulting in the creation of connectome datasets of unprecedented size. MaPPeRTrac is publicly available and tested on commercial and scientific hardware, so it can accelerate brain connectome research for a broader user community. MaPPeRTrac is available at: https://github.com/LLNL/mappertrac .

Changes in the Networks of Attention across the Lifespan: A Graphical Meta-Analysis

Three Posnerian networks of attention (alerting, orienting, and executive control) have been distinguished on the bases of behavioural, neuropsychological, and neuroscientific evidence. Here, we examined the trajectories of these networks throughout the human lifespan using the various Attention Network Tests (ANTs), which were specifically developed to measure the efficacy of these networks. The ANT Database was used to identify relevant research, resulting in the inclusion of 36 publications. We conducted a graphical meta-analysis using network scores from each study, based on reaction time plotted as a function of age group. Evaluation of attentional networks from childhood to early adulthood suggests that the alerting network develops relatively quickly, and reaches near-adult level by the age of 12. The developmental pattern of the orienting network seems to depend on the information value of the spatial cues. Executive control network scores show a consistent decrease (improvement) with age in childhood. During adulthood (ages 19-75), changes in alerting depend on the modality of the warning signal, while a moderate increase in orienting scores was seen with increasing age. Whereas executive control scores, as measured in reaction time, increase (deterioration) from young adulthood into later adulthood an opposite trend is seen when scores are based on error rates.

Extracting the most discriminating functional connections in mild traumatic brain injury based on machine learning

BACKGROUND

Mild traumatic brain injury (mTBI) is characterized as brain microstructural damage, which may cause a wide range of brain functional disturbances and emotional problems. Brain network analysis based on machine learning is an important means of neuroimaging research. Obtaining the most discriminating functional connection is of great significance to analyze the pathological mechanism of mTBI.

METHODS

To better obtain the most discriminating features of functional connection networks, this study proposes a hierarchical feature selection pipeline (HFSP) composed of Variance Filtering (VF), Lasso, and Principal Component Analysis (PCA). Ablation experiments indicate that each module plays a positive role in classification, validating the robustness and reliability of the HFSP. Furthermore, the HFSP is compared with recursive feature elimination (RFE), elastic net (EN), and locally linear embedding (LLE), verifying its superiority. In addition, this study also utilizes random forest (RF), SVM, Bayesian, linear discriminant analysis (LDA), and logistic regression (LR) as classifiers to evaluate the generalizability of HFSP.

RESULTS

The results show that the indexes obtained from RF are the highest, with accuracy = 89.74%, precision = 91.26%, recall = 89.74%, and F1 score = 89.42%. The HFSP selects 25 pairs of the most discriminating functional connections, mainly distributed in the frontal lobe, occipital lobe, and cerebellum. Nine brain regions show the largest node degree.

LIMITATIONS

The number of samples is small. This study only includes acute mTBI.

CONCLUSIONS

The HFSP is a useful tool for extracting discriminating functional connections and may contribute to the diagnostic processes.

Altered longitudinal structural connectome in paediatric mild traumatic brain injury: an Advancing Concussion Assessment in Paediatrics study

Advanced diffusion-weighted imaging techniques have increased understanding of the neuropathology of paediatric mild traumatic brain injury (i.e. concussion). Most studies have examined discrete white-matter pathways, which may not capture the characteristically subtle, diffuse and heterogenous effects of paediatric concussion on brain microstructure. This study compared the structural connectome of children with concussion to those with mild orthopaedic injury to determine whether network metrics and their trajectories across time post-injury differentiate paediatric concussion from mild traumatic injury more generally. Data were drawn from of a large study of outcomes in paediatric concussion. Children aged 8-16.99 years were recruited from five paediatric emergency departments within 48 h of sustaining a concussion (n = 360; 56% male) or mild orthopaedic injury (n = 196; 62% male). A reliable change score was used to classify children with concussion into two groups: concussion with or without persistent symptoms. Children completed 3 T MRI at post-acute (2-33 days) and/or chronic (3 or 6 months, via random assignment) post-injury follow-ups. Diffusion-weighted images were used to calculate the diffusion tensor, conduct deterministic whole-brain fibre tractography and compute connectivity matrices in native (diffusion) space for 90 supratentorial regions. Weighted adjacency matrices were constructed using average fractional anisotropy and used to calculate global and local (regional) graph theory metrics. Linear mixed effects modelling was performed to compare groups, correcting for multiple comparisons. Groups did not differ in global network metrics. However, the clustering coefficient, betweenness centrality and efficiency of the insula, cingulate, parietal, occipital and subcortical regions differed among groups, with differences moderated by time (days) post-injury, biological sex and age at time of injury. Post-acute differences were minimal, whereas more robust alterations emerged at 3 and especially 6 months in children with concussion with persistent symptoms, albeit differently by sex and age. In the largest neuroimaging study to date, post-acute regional network metrics distinguished concussion from mild orthopaedic injury and predicted symptom recovery 1-month post-injury. Regional network parameters alterations were more robust and widespread at chronic timepoints than post-acutely after concussion. Results suggest that increased regional and local subnetwork segregation (modularity) and inefficiency occurs across time after concussion, emerging after post-concussive symptom resolve in most children. These differences persist up to 6 months after concussion, especially in children who showed persistent symptoms. While prognostic, the small to modest effect size of group differences and the moderating effects of sex likely would preclude effective clinical application in individual patients.

Predicting sex, age, general cognition and mental health with machine learning on brain structural connectomes

There is an increasing expectation that advanced, computationally expensive machine learning (ML) techniques, when applied to large population-wide neuroimaging datasets, will help to uncover key differences in the human brain in health and disease. We take a comprehensive approach to explore how multiple aspects of brain structural connectivity can predict sex, age, general cognitive function and general psychopathology, testing different ML algorithms from deep learning (DL) model (BrainNetCNN) to classical ML methods. We modelled N = 8183 structural connectomes from UK Biobank using six different structural network weightings obtained from diffusion MRI. Streamline count generally provided the highest prediction accuracies in all prediction tasks. DL did not improve on prediction accuracies from simpler linear models. Further, high correlations between gradient attribution coefficients from DL and model coefficients from linear models suggested the models ranked the importance of features in similar ways, which indirectly suggested the similarity in models' strategies for making predictive decision to some extent. This highlights that model complexity is unlikely to improve detection of associations between structural connectomes and complex phenotypes with the current sample size.

Abnormal structural and functional network topological properties associated with left prefrontal, parietal, and occipital cortices significantly predict childhood TBI-related attention deficits: A semi-supervised deep learning study

Introduction

Traumatic brain injury (TBI) is a major public health concern in children. Children with TBI have elevated risk in developing attention deficits. Existing studies have found that structural and functional alterations in multiple brain regions were linked to TBI-related attention deficits in children. Most of these existing studies have utilized conventional parametric models for group comparisons, which have limited capacity in dealing with large-scale and high dimensional neuroimaging measures that have unknown nonlinear relationships. Nevertheless, none of these existing findings have been successfully implemented to clinical practice for guiding diagnoses and interventions of TBI-related attention problems. Machine learning techniques, especially deep learning techniques, are able to handle the multi-dimensional and nonlinear information to generate more robust predictions. Therefore, the current research proposed to construct a deep learning model, semi-supervised autoencoder, to investigate the topological alterations in both structural and functional brain networks in children with TBI and their predictive power for post-TBI attention deficits.

Methods

Functional magnetic resonance imaging data during sustained attention processing task and diffusion tensor imaging data from 110 subjects (55 children with TBI and 55 group-matched controls) were used to construct the functional and structural brain networks, respectively. A total of 60 topological properties were selected as brain features for building the model.

Results

The model was able to differentiate children with TBI and controls with an average accuracy of 82.86%. Functional and structural nodal topological properties associated with left frontal, inferior temporal, postcentral, and medial occipitotemporal regions served as the most important brain features for accurate classification of the two subject groups. Post hoc regression-based machine learning analyses in the whole study sample showed that among these most important neuroimaging features, those associated with left postcentral area, superior frontal region, and medial occipitotemporal regions had significant value for predicting the elevated inattentive and hyperactive/impulsive symptoms.

Discussion

Findings of this study suggested that deep learning techniques may have the potential to help identifying robust neurobiological markers for post-TBI attention deficits; and the left superior frontal, postcentral, and medial occipitotemporal regions may serve as reliable targets for diagnosis and interventions of TBI-related attention problems in children.

Edge Density Imaging Identifies White Matter Biomarkers of Late-Life Obesity and Cognition

Alzheimer disease (AD) and obesity are related to disruptions in the white matter (WM) connectome. We examined the link between the WM connectome and obesity and AD through edge-density imaging/index (EDI), a tractography-based method that characterizes the anatomical embedding of tractography connections. A total of 60 participants, 30 known to convert from normal cognition or mild-cognitive impairment to AD within a minimum of 24 months of follow up, were selected from the Alzheimer disease Neuroimaging Initiative (ADNI). Diffusion-weighted MR images from the baseline scans were used to extract fractional anisotropy (FA) and EDI maps that were subsequently averaged using deterministic WM tractography based on the Desikan-Killiany atlas. Multiple linear and logistic regression analysis were used to identify the weighted sum of tract-specific FA or EDI indices that maximized correlation to body-mass-index (BMI) or conversion to AD. Participants from the Open Access Series of Imaging Studies (OASIS) were used as an independent validation for the BMI findings. The edge-density rich, periventricular, commissural and projection fibers were among the most important WM tracts linking BMI to FA as well as to EDI. WM fibers that contributed significantly to the regression model related to BMI overlapped with those that predicted conversion; specifically in the frontopontine, corticostriatal, and optic radiation pathways. These results were replicated by testing the tract-specific coefficients found using ADNI in the OASIS-4 dataset. WM mapping with EDI enables identification of an abnormal connectome implicated in both obesity and conversion to AD.

Safeguarding Athletes Against Head Injuries Through Advances in Technology: A Scoping Review of the Uses of Machine Learning in the Management of Sports-Related Concussion

Sports injury prevention is an important part of the athlete welfare and safeguarding research field. In sports injury prevention, sport-related concussion (SRC) has proved to be one of the most difficult and complex injuries to manage in terms of prevention, diagnosis, classification, treatment and rehabilitation. SRC can cause long-term health issues and is a commonly reported injury in both adult and youth athletes around the world. Despite increased knowledge of the prevalence of SRC, very few tools are available for diagnosing SRC in athletic settings. Recent technological innovations have resulted in different machine learning and deep learning methodologies being tested to improve the management of this complex sports injury. The purpose of this article is to summarize and map the existing research literature on the use of machine learning in the management of SRC, ascertain where there are gaps in the existing research and identify recommendations for future research. This is explored through a scoping review. A systematic search in the three electronic databases SPORTDiscus, PubMed and Scopus identified an initial 522 studies, of which 24 were included in the final review, the majority of which focused on machine learning for the prediction and prevention of SRC (N = 10), or machine learning for the diagnosis and classification of SRC (N = 11). Only 3 studies explored machine learning approaches for the treatment and rehabilitation of SRC. A main finding is that current research highlights promising practical uses (e.g., more accurate and rapid injury assessment or return-to-sport participation criteria) of machine learning in the management of SRC. The review also revealed a narrow research focus in the existing literature. As current research is primarily conducted on male adolescents or adults from team sports in North America there is an urgent need to include wider demographics in more diverse samples and sports contexts in the machine learning algorithms. If research datasets continue to be based on narrow samples of athletes, the development of any new diagnostic and predictive tools for SRC emerging from this research will be at risk. Today, these risks appear to mainly affect the health and safety of female athletes.

The role of artificial intelligence in paediatric neuroradiology

Imaging plays a fundamental role in the managing childhood neurologic, neurosurgical and neuro-oncological disease. Employing multi-parametric MRI techniques, such as spectroscopy and diffusion- and perfusion-weighted imaging, to the radiophenotyping of neuroradiologic conditions is becoming increasingly prevalent, particularly with radiogenomic analyses correlating imaging characteristics with molecular biomarkers of disease. However, integration into routine clinical practice remains elusive. With modern multi-parametric MRI now providing additional data beyond anatomy, informing on histology, biology and physiology, such metric-rich information can present as information overload to the treating radiologist and, as such, information relevant to an individual case can become lost. Artificial intelligence techniques are capable of modelling the vast radiologic, biological and clinical datasets that accompany childhood neurologic disease, such that this information can become incorporated in upfront prognostic modelling systems, with artificial intelligence techniques providing a plausible approach to this solution. This review examines machine learning approaches than can be used to underpin such artificial intelligence applications, with exemplars for each machine learning approach from the world literature. Then, within the specific use case of paediatric neuro-oncology, we examine the potential future contribution for such artificial intelligence machine learning techniques to offer solutions for patient care in the form of decision support systems, potentially enabling personalised medicine within this domain of paediatric radiologic practice.

Structural connectome differences in pediatric mild traumatic brain and orthopedic injury

Sophisticated network‐based approaches such as structural connectomics may help to detect a biomarker of mild traumatic brain injury (mTBI) in children. This study compared the structural connectome of children with mTBI or mild orthopedic injury (OI) to that of typically developing (TD) children. Children aged 8–16.99 years with mTBI (n = 83) or OI (n = 37) were recruited from the emergency department and completed 3T diffusion MRI 2–20 days postinjury. TD children (n = 39) were recruited from the community and completed diffusion MRI. Graph theory metrics were calculated for the binarized average fractional anisotropy among 90 regions. Multivariable linear regression and linear mixed effects models were used to compare groups, with covariates age, hemisphere, and sex, correcting for multiple comparisons. The two injury groups did not differ on graph theory metrics, but both differed from TD children in global metrics (local network efficiency: TD > OI, mTBI, d = 0.49; clustering coefficient: TD < OI, mTBI, d = 0.49) and regional metrics for the fusiform gyrus (lower degree centrality and nodal efficiency: TD > OI, mTBI, d = 0.80 to 0.96; characteristic path length: TD < OI, mTBI, d = −0.75 to −0.90) and in the superior and middle orbital frontal gyrus, paracentral lobule, insula, and thalamus (clustering coefficient: TD > OI, mTBI, d = 0.66 to 0.68). Both mTBI and OI demonstrated reduced global and regional network efficiency and segregation as compared to TD children. Findings suggest a general effect of childhood injury that could reflect pre‐ and postinjury factors that can alter brain structure. An OI group provides a more conservative comparison group than TD children for structural neuroimaging research in pediatric mTBI.