Use this URL to cite or link to this record in EThOS:
Title: Point process modelling of coordinate-based meta-analysis neuroimaging data
Author: Samartsidis, Pantelis
ISNI:       0000 0004 6062 0930
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Now over 25 years old, functional magnetic resonance imaging (fMRI) has made significant contributions in improving our understanding of the human brain function. However, some limitations of fMRI studies, including those associated with the small sample sizes that are typically employed, raise concerns about validity of the technique. Lately, growing interest has been observed in combining the results of multiple fMRI studies in a meta-analysis. This can potentially address the limitations of single experiments and raise opportunities for reaching safer conclusions. Coordinate-based meta-analyses (CBMA) use the peak activation locations from multiple studies to find areas of consistent activations across experiments. CBMA presents statisticians with many interesting challenges. Several issues have been solved but there are also many open problems. In this thesis, we review literature on the topic and after describing the unsolved problems we then attempt to address some of the most important. The first problem that we approach is the incorporation of study-specific characteristics in the meta-analysis model known as meta-regression. We propose an novel meta-regression model based on log-Gaussian Cox processes and develop a parameter estimation algorithm using the Hamiltonian Monte Carlo method. The second problem that we address is the use of CBMA data as prior in small underpowered fMRI studies. Based on some existing work on the topic, we develop a hierarchical model for fMRI studies that uses previous CBMA findings as a prior for the location of the effects. Finally, we discuss a classical problem of meta-analysis, the file drawer problem, where studies are suppressed from the literature because they fail to report any significant finding. We use truncated models to infer the total number of non-significant studies that are missing from a database. All our methods are tested on both simulated and real data.
Supervisor: Not available Sponsor: National Institutes of Health (U.S.)
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: RC Internal medicine