Robust Methods for Neuroimaging Meta Analyses

external supervisor

Stephen M Smith, Nuffield Department of Clinical Neurosciences

BACKGROUND

Meta-analysis (MA) is essential for combining research results and identifying the most likely effect magnitude and quantifying its uncertainty.  In brain imaging, neuroimaging meta-analysis (NMA) has additional challenges of working with 3D volume data, having very heterogenous collections of studies, and having varying number of repeated results per study. In this project you will develop and assess several methods to make NMA valid and reliable.

RESEARCH EXPERIENCE, RESEARCH METHODS AND TRAINING

The work will start with a comprehensive literature review on standard (non-imaging) meta-analysis focusing on robust estimation and inference methods, techniques for increasing validity with small numbers of studies, and methods specific to standardised effect sizes used in neuroimaging. This will lead to the first project, where existing random effects meta-analysis methods will be combined with spatial regularisation to lower the number of studies needed for validity and stability. The next project will focus on robust methods for standardised effect sizes to provide trustworthy results even when a proportion of the studies are corrupted or irrelevant for some reason, drawing on recent statistical methods developed for Mendelian Randomisation. A final project will consider the challenge of varying number of MA inputs per study, evaluating traditional repeated measures models and novel robust combining methods.

FIELD WORK, SECONDMENTS, INDUSTRY PLACEMENTS AND TRAINING

Training will be provided in neuroimaging via the WIN MRI Graduate Programme, an intensive 2-term programme to familiarise the student with magnetic resonance imaging acquisition and analysis.

PROSPECTIVE  STUDENT

The ideal candidate will have a Master’s degree in statistics or other quantitative area; experience with meta-analysis or brain imaging data is a plus but not essential.