Background. Traumatic Brain Injury (TBI) is a debilitating condition, with long-term sequelae, affecting a considerable portion of the population. Typically, TBI transpires from a direct impact to the head as well as from the onset of acceleration and deceleration forces following abrupt changes in head position. Both localized and diffuse damage generated after injury are subject to a cascade of events altering normal biological function, culminating in heterogeneous pathophysiology. Considerable attempts to tackle this heterogeneity have been made. However, this has lead to numerous TBI classifications primarily rooted in subjective measurements, such as the Glasgow Coma Scale score, loss of consciousness duration, and post-traumatic amnesia, each known to be poor indicators of TBI. Physical measures, such as clinical findings from Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) scans, are indispensable for assessing TBI, but they lack reliability and specificity in diagnosis and prognosis of TBI. Part of this deficiency arises from the reliance on subjective interpretation of these scans. A more significant issue arises as a result of inadequate methods that are unable to capture the extent and heterogeneity of damage. Objective. To extensively capture the heterogeneous pathophysiology in TBI using objective and physical measurements to advance candidate neuroimaging biomarkers for TBI, which will aid in overcoming the critical barriers that currently exist in diagnosis and prognosis of TBI. Contributions. For my first contribution, I developed and validated an algorithm that incorporates natural language processing methods to automate the extraction of TBI clinical findings from head CT scan radiology reports. The output of my algorithm is a structured summary of twenty-seven standardized TBI clinical findings with their respective status. My algorithm was validated using physician annotators as the gold standard and was found to be equivalent, sensitive, specific, and fast. For my second contribution, I examined the shape of the diffusion tensor, explored microstructural disturbances using voxelwise tract-based spatial statistics, quantified the location of these disturbances, and established an ordered ranking of diffusion measurements using machine learning methods. I discovered widespread changes in diffusion measurements, often affecting long-range connections, indicating disturbances in microstructural integrity after TBI. Lastly, I reported that structural diffusion measurements were better able to differentiate TBI from control groups. For my third contribution, I evaluated the construction of TBI structural networks using spectral graph theory and statistical network analysis to identify the most indicative diffusion measurements representing disrupted microarchitecture. I set out to uncover distinctions between diffusion measurements used in the construction of TBI structural networks, locate network disruptions, and discover preliminary network topologies. Using vastly different methodologies from my second contribution, I again found that structural diffusion measurements were better able to differentiate TBI from control groups. In addition, I discovered disrupted connections were primarily located in the parietal and temporal lobes, with geometric diffusion measurements more prevalent in these disruptions. These results indicate a dual representation of microarchitectural disruption in TBI structural networks. For my fourth contribution, I assessed TBI structural and functional networks using graph theory. I examined five measures of network topology: construction, integration, segregation, influence, and resilience. I discovered altered network integration and network influence in TBI structural networks, in addition to altered network resilience in TBI structural and functional networks, with no alterations to network segregation. My results indicate altered information exchange capacity, participation capacity, and flexibility of the TBI brain network. Discussion. The nearly universal acquisition of head CT scans following trauma necessitates the inclusion of their findings in discerning injuries. These head CT scans are often accompanied by a radiology report containing expert-level interpretation in unstructured narrative text format, which makes retrieving information challenging. The outcome of my first contribution addresses this and can be used to partition patients into meaningful groups based on physical measurements. A notable portion of TBI patients do not have clinical findings on hospital admission head CT scans yet display pathologies on MRI consistent with microstructural damage. Efforts to capture this damage have been extensive. However, there are discrepancies in identifying the location of microstructural damage and the direction of changes in measurements quantifying it. Part of this is due to an incomplete assessment of the diffusion tensor shape. The outcome of my second contribution addresses this deficiency, provides an assessment of underlying pathophysiological mechanisms associated with microstructural disturbances, and postulates candidate predictors of microstructural damage in TBI. The construction and study of connectomes, a field known as connectomics, provides a theoretical framework for understanding the brain as a complex network. Since microstructural disturbances in TBI lead to disruptions to the structural network, quantifying these disruptions are crucial for describing TBI pathophysiology, yet this quantification has not yet been previously tackled. The outcome of my third contribution addresses this and establishes more suitable methods for TBI structural network construction, which may aid in the precise identification of vulnerable brain regions. Quantification of diffuse injury over the entire structural network is required to appreciate network dysfunction. Naturally, this can be addressed by assessing alterations to connectome topology. The outcome of my fourth contribution addresses this and establishes the impact of alterations to network topology in TBI, which may lead to improvements in patient monitoring by assessing the progression of these topological alterations towards standardized levels. Impact. In summary, my dissertation is uniquely positioned to create valuable contributions towards identifying neuroimaging biomarkers that will catalyze research on TBI. I expect my contributions will lead to a significant shift in current TBI research by adopting novel network analysis and machine learning methods that go beyond conventional approaches and ultimately improve the lives of the TBI-afflicted.
University of Minnesota Ph.D. dissertation. May 2019. Major: Biomedical Informatics and Computational Biology. Advisor: Uzma Samadani. 1 computer file (PDF); xvi, 287 pages.
Uncovering Disturbed Microstructure, Disrupted Microarchitecture, and Altered Network Topology in Traumatic Brain Injury.
Retrieved from the University of Minnesota Digital Conservancy,
Content distributed via the University of Minnesota's Digital Conservancy may be subject to additional license and use restrictions applied by the depositor.