Brain Connectivity Modeling in Soldiers with Mild-Traumatic Brain Injury and Posttraumatic Stress Disorder
Type of DegreePhD Dissertation
DepartmentElectrical and Computer Engineering
MetadataShow full item record
Functional magnetic resonance imaging (fMRI) has been increasingly used for understanding cognitive processes in both healthy and clinical populations. In this work, we employed various computational approaches in studying brain alterations in U.S. Army soldiers with posttraumatic stress disorder (PTSD) and mild-traumatic brain injury (mTBI). PTSD and mTBI share largely similar symptoms, and have high comorbidity in military populations, with 7% of war veterans acquiring both disorders. Despite such high prevalence, the neural underpinnings of PTSD as well as comorbid PTSD and mTBI remain poorly understood. We employ multiple approaches, including functional connectivity (FC) modeling, effective connectivity (EC) modeling and complex network analysis, to investigate brain disruptions in these disorders. Notably, we used dynamic connectivity in all our analyses for characterizing variability of connectivity over time, in addition to traditionally used static connectivity measures. Using resting-state fMRI, we first employed static and time-varying FC to identify significantly altered co-activation patterns which exhibit temporally “frozen” hyper-connected profile in the disorders. Using whole-brain connectivity in a data-driven manner without the imposition of any priors or assumptions, we identified only the hippocampus-striatum path to be significantly altered in the disorders, which likely represents habit-like response associated with traumatic memories. This path also had high behavioral relevance. Using machine learning classification, we showed that this path is a potential imaging biomarker of PTSD and mTBI. Next, we performed EC analysis using Granger causality for identifying sources of network disturbances in PTSD and mTBI. Causality, or directional connectivity quantified using EC, is characteristically different from co-activation or FC. While EC gives connections between regions, the source of disruption should be a region, not connection. We thus employed a probabilistic framework to identify the source region(s) of disruption using a novel framework. We found that the middle frontal gyrus is the source of disruptions, whose dysregulation causes overdrive in subcortical regions, leading to heightened emotional response to traumatic memories. The identified paths also had high behavioral relevance and diagnostic ability. Though the results obtained from EC analyses were informative, we recognized that connectivity modeling of individual paths does not capture alterations in network architecture which are critical for producing complex behaviors. We thus employed complex network analysis to study network-level alterations in the disorders using EC networks. We studied specialized processing (segregation) and efficient communication (integration), as well as their variability using time-varying network dynamics technique developed by us. We identified network-level markers which help in distinguishing between PTSD and comorbid conditions, a vexed problem whose solution has been elusive in current literature. We found alterations of network architecture in two sub-networks, fronto-visual and fronto-subcortical, with disruption primarily originating in prefrontal areas of cognitive control. Taken together, FC and EC analyses provided novel insights into the underlying network structure, the flow of information and the foci of disruption in these disorders, which might help in developing objective diagnosis and treatments for these disorders. Next, using fMRI during an emotion regulation task, we studied the network of cognitive emotion regulation in healthy adults and its dysregulation in comorbid PTSD and mTBI. This was important since emotion dysregulation is seen as the major cause of symptoms observed in disorders like these. We identified activated regions using GLM analysis. We performed EC analysis using the timeseries obtained from the identified regions, which provided the network of emotion regulation and dysregulation while actively engaged in an emotion regulation task as opposed to resting-state. With all the aforementioned analyses, one aspect which needs mentioning is that fMRI is not a direct measure of neural activity. It measures blood oxygenation which is an indirect measure, hence susceptible to sources of variability which are non-neural in origin. The transfer function between neural activity and fMRI, called the hemodynamic response function (HRF), is known to vary across the brain in the same subject, and across subjects. We obtained the HRF at every voxel using blind deconvolution. We hypothesized that there are group-wise differences in HRF and that they may drive connectivity differences if HRF variability is not removed from the data. We found significant HRF differences between the groups mainly in posterior cingulate, precuneus and secondary visual areas. We performed seed-based connectivity using them, and found that ignoring HRF variability during connectivity analysis leads to possible false positive and false negative connectivities. In summary, we propose and test a comprehensive mechanistic model of brain alterations in soldiers with PTSD and mTBI, and illustrate the precautions to be followed during fMRI analysis to reliably characterize brain functioning in these disorders. We hope that this work contributes towards the development of effective diagnoses and treatments for PTSD and mTBI. Finally, the tenets of the proposed analyses framework is agnostic and generally applicable for characterizing brain alterations underlying various mental disorders and cognitive domains using neuroimaging.