Global Outcome and Cognitive Control Function in Moderate-to-Severe Traumatic Brain Injury - Classification and Prediction

Abstract

Previous research has shown a close relationship between unfavourable global outcome and cognitive control dysfunction in moderate-to-severe traumatic brain injury. Proponents of graph theory frequently proclaim its possible utility for diagnosis and the investigation of clinical phenomena. In this thesis, we evaluate the performance of a graph theory-based classification model for global outcomes in patients with moderate-to-severe traumatic brain injury. Moreover, we evaluate a four-class classification model for performance-based cognitive control dysfunction in both patients and healthy controls. To investigate the multifaceted nature of cognitive control function, the relationship between networks of the brain and self-reported cognitive control function is also explored. Estimation of individual graph metrics for 68 patients with moderate-to-severe traumatic brain injury, along with 68 healthy controls, were derived from resting state fMRI data. The participants also underwent neuropsychological evaluation of global outcomes and cognitive control function. All classification models performed significantly better than randomized models. However, classification models trained with graph metrics performed significantly better than did models only using partial correlation between regions of interest in the four-class model of performance-based cognitive control function. Given the simplicity and interpretability of the model, this difference in performance suggests that graph theory is appropriate only for a subgroup of classification problems. The models’ optimal performance using a variety of graph metrics implies that a combination of different graph metrics best describes the clinical phenomena of global outcome and cognitive control function, reflecting a redistribution of network hubs and coinciding hyper- and hypoconnectivity. In addition, the anatomical location of the most informative graph metrics shows how a neurocognitive model of cognitive control function must take into consideration a network perspective that goes beyond a “core network” of cognitive control function comprising dorsal medial prefrontal cortex, insula, paracingulate gyrus, supramarginal gyrus, and middle temporal gyrus. Supporting recent findings concerning task-fMRI, we also show a relationship between functional connectivity and self-reported cognitive control function.