Connectome-based predictive modelling of neurocognitive and affective processes

Abstract

Brain-based prediction of individual variations in cognitive and affective processes can provide deep and valuable insight into the brain mechanisms underlying those processes, and is useful for modelling individual patient’s functions and behaviours in clinical contexts. Despite recent efforts to predict human behaviour using brain-imaging data and machine-learning methods, there are still significant gaps in this area. These relate to the neural features identified in existing studies that explore brain–behaviour relationships, which may not show high generalisability to other populations and are often based on group differences rather than individual-level prediction results. Moreover, there has been little research into the prediction of important neurocognitive functions, such as processing speed (PS) and behaviours related to affective dysregulation, such as suicide, among healthy and clinical populations. This thesis aims to address these gaps and describes three studies aimed at predicting PS and suicidality among healthy and clinical populations, using a recently developed connectome-based predictive modelling (CPM) approach. Study 1 (Chapter 2) explores whether resting-state functional connectivity (rs-FC) can be used to predict PS in cognitively healthy older adults. The results showed that the CPM models can significantly predict PS in this age group, as well as attention and memory. The PS-predictive models can be generalised to predict PS in an independent sample of older adults, but not to samples of younger or middle-aged adults. These findings correspond with previous studies that identified close associations between PS, attention and memory, and support the application of CPM in the prediction of age-related changes in neurocognitive functioning. Study 2 (Chapter 3) extended the prediction of PS to a specific clinical population—patients with moyamoya disease (MMD). The PS of MMD patients’, 1 to 6 months after receiving surgery for MMD, was significantly predicted by a CPM model based on their pre-surgical rs-FC patterns. These results consolidated previous findings that MMD is closely associated with PS deficits. The findings have potential for predicting post-operative neurocognitive outcomes in MMD patients, which will help inform clinical decision making and patient management. Study 3 (Chapter 4) further explored the role of rs-FC in predicting neurocognitive functioning in healthy and clinical samples, focusing on the use of structural and rs-FC patterns in patients with affective dysregulation, including suicide—a lethal behaviour that is highly prevalent among patients with major depressive disorder. A sample of older adults with late-life depression (LLD) was studied using various assessment tools to determine whether CPM brain-connectivity models could predict suicide risk. Results showed that both rs-FC and structural-connectivity models can predict the severity of suicide risk in LLD patients. This thesis provides compelling evidence for the use of brain-connectivity data to build connectome-based models capable of predicting individual differences in neurocognitive and affective processes, and further advances our understanding of the brain-network mechanisms underpinning important dysregulation processes. Taken together, the results have the potential to improve clinical diagnoses, prognosis and patient management, and inform brain-targeting interventions.