Hierarchical Bayesian Models for Multimodal Neuroimaging Data
Doctor of Philosophy
Within the past few decades, advances in imaging acquisition have given rise to a large number of in vivo techniques for brain mapping. This wide range of structural and functional imaging modalities provides a major source of information-rich data which may be used to non-invasively understand the human brain, and is a promising source of information for improved clinical diagnosis and treatment decision-making. Due to the complex spatial structure of neuroimaging data as well as the small number of samples typically collected in neuroimaging experiments, statistical methods which try to integrate different types of neuroimaging data are paramount. Our research is focused on the development of methods which allow for incorporation of prior information from multimodal neuroimaging sources to improve the reliability of inference in the presence of small to moderate sample sizes. First, we propose an integrative predictive modeling framework for neuroimaging data with spatial structure, such as positron emission tomography or structural MRI. The method provides a unified framework for the identification of pathologic subgroups, identification of imaging biomarkers characterizing the pathologic states, and prediction of the clinical outcome of interest. Furthermore, Bayesian priors are used to inform the selection of imaging markers with external imaging data. We assess the performance of our method on synthetic data and compare its performance to competing methods. We demonstrate use of the proposed method for identifying markers for patients at high-risk for poor treatment outcome from a study on temporal lobe epilepsy patients undergoing anterior temporal lobe resection. Second, we propose a multi-subject approach for effective connectivity inference using resting-state functional MRI data. The proposed method provides a joint, single-stage framework for multi-subject effective connectivity inference at both the subject- and group-levels. Furthermore, it accounts for multi-modal imaging data by integrating structural imaging information into the prior model. We investigate the performance of the proposed model on simulated data, and demonstrate through simulation studies that the approach results in improved inference on effective connectivity at both the subject- and group-levels compared to currently used methods. The proposed method is illustrated through an application to resting-state functional MRI and structural MRI for identifying effective connectivity in temporal lobe epilepsy patients and healthy controls.
Bayesian hierarchical model; Positron emission tomography (PET); Functional magnetic resonance imaging (fMRI); Structural MRI; Spatially-informed prior; More... Mixture model; Variable selection; Polya-Gamma distribution; Vector Autoregressive (VAR) model; Granger Causality; Effective Connectivity; Multimodal Neuroimaging; Temporal Lobe Epilepsy Less...