Exploring the large-scale mechanisms orchestrating human brain functioning through dynamic causal modelling of resting-state fMRI

The quest to unravel the intricate mechanisms governing the multiple brain functionalities has spurred the development of various computational and theoretical approaches in brain information processing. These endeavours aim to elucidate the mechanisms underpinning the multiscale organization of the brain. Notably, contemporary human neuroscience research has increasingly focused on statistical approaches, particularly in analysing functional magnetic resonance imaging (fMRI) during the resting state. However, establishing a direct link between these statistical methods and plausible underlying neural mechanisms proves challenging due to their static, local, and inferential nature. Consequently, this study advocates for the adoption of computational tools from dynamical systems theory, providing a crucial mechanistic framework for characterizing the brain's time-varying dynamics. In this thesis, we adopted a dynamic causal modeling (DCM) framework to deduce subject-specific parameters outlining the fMRI dynamics within an individual's state space. We systematically analyzed these parameters to explore the overarching generative mechanisms that link cerebral activity, blood flow, and metabolism on a large scale, whose intricate interplay mediates crucial communication processes integral to human cognition. Our exploration commenced with an examination of neurovascular coupling, shaped by the hemodynamic response function, analysing its spatial variability across the cortex and subcortex across the human lifespan and then assessing the predictive properties of vascular features through a multivariate classification model predicting the age class of each individual. Subsequently, we proposed a dynamic treatment of the effective connectome through its dynamic mode decomposition, partitioning the corresponding spectral domain based on energetic measures of the emergent kinetic flow. This dynamic approach provided insights into hierarchical and heterarchical flows governing spatiotemporal brain dynamics, emphasizing their contributions to subject-specific information and the efficiency of cognitive processes. These findings were encapsulated through specific metrics that measured subject identifiability and the multivariate correlation with cognitive scores, thereby illuminating the complex interconnection between brain dynamics and cognitive functioning. Ultimately, we delved into the realm of neurometabolic coupling, exploring conceivable low-dimensional associations between dynamic flows and metrics of brain glucose consumption through the utilization of a partial least squares correlation framework. Moreover, we assessed whether complementary information from such effective-metabolic coupling could enhance our understanding of how observed impairments contribute to the mechanisms underlying glioma progression. In summary, the contributions of this dissertation affirm the powerful role of the generative, dynamic, and mechanistic DCM framework in enriching our understanding of fundamental large-scale vascular, neural, and metabolic mechanisms, elucidating their crucial role in shaping human brain functioning. Despite limitations imposed by linearized dynamics, the proposed treatment of such modeling approach represents an initial step towards a more dynamic and mechanistic interpretation of DCM parameters. This approach, extending beyond the mere statistical fitting of data, unveils its potential to revolutionize neuroimaging data analysis, facilitating the identification of large-scale phenomena characterizing both healthy and pathological states.