FMRIB Analysis - Research & Resources
Below we give a brief summary of research areas covered by the FMRIB Analysis Group, and some research-related resources that we have generated.
We have worked on (univariate) time series analysis (first-level FMRI session analysis) and Bayesian mixed-effects multi-session/multi-subject analysis. These projects have provided the core statistics for FEAT - an advanced FMRI analysis tool with easy-to-use yet powerful GUI for complete first- and higher-level FMRI analysis. We are also working on optimal (haemodynamic response function) basis functions and related estimation and other more advanced Bayesian spatiotemporal signal and noise modelling.
We have developed probabilistic ICA (independent component analysis) for model-free FMRI analysis, for both timeseries and multi-subject analyses. This has resulted in the MELODIC tool. More recently, we have developed an approach for allowing the comparison of ICA components (e.g. RSNs) across subjects, based on group-ICA followed by dual regression.
We have been working on network modelling from FMRI timeseries, and have developed network simulations in order to evaluate a wide range of network modelling methods.
There are several studies where we have generated RSN and TFM spatial maps, for example for use as templates to compare against new studies. This includes:
We have developed a number of robust tools for the analysis of structural MRI data. SUSAN is a tool for nonlinear noise reduction. BET segments brain from non-brain in structural and functional data. FAST segments a brain image into different tissue types, and provides bias field correction. FLIRT gives robust affine (linear) inter- and intra-modal registration, and FNIRT gives robust nonlinear registration. SIENA is a fully automated tool for estimating structural brain state and temporal change (e.g. for estimating brain atrophy).
We have developed new methods for analysing diffusion imaging data. FDT - FMRIB's Diffusion Toolbox - includes diffusion tensor fitting (including FA estimation), Bayesian fibre tract direction analysis and probabilistic tractography. Also part of this is TBSS - Tract-Based Spatial Statistics; this is an approach to carrying out voxel-based multi-subject statistics on diffusion (e.g. FA) data.
MRI Physics Modelling
We are also working on MR analysis research that is closely tied in with understanding the MR physics behind the data. FUGUE is a tool for unwarping EPI data on the basis of a B0 fieldmap. POSSUM is a project developing a full MRI data simulator from the basic physics, allowing the study of (e.g.) various MRI and FMRI artefacts.
We have worked on spatial mixture modelling - a way of modelling a mixture of distributions (activation-nonactivation, different tissue types, etc.) with adaptive spatial classification regularisation, for use as an inference tool in FMRI etc.
We are also working on cluster-like thresholding (Threshold-Free Cluster Enhancement), without having to define an initial cluster-forming threshold.