Methods research at the MRC CBU is distributed across several groups and projects.
We are developing new methods in the following broad areas:
- Physics of magnetic resonance imaging
- Analysis and modelling of (f)MRI data
- Analysis and modelling of EEG/MEG data
- Statistical modelling
Some of the most active areas of development are listed below:
Multimodal imaging – Rik Henson
We are continuing to refine methods for combining neuroimaging data from multiple modalities, such as MEG and EEG (Henson et al., 2009b), and M/EEG and fMRI (Henson et al., 2010), in order to optimise both temporal and spatial resolution. For this, we use a parametric empirical Bayesian framework (reviewed in Henson et al., 2011). These methods are incorporated into the SPM software (Litvak et al, 2011), and accompanied by a multisubject, multimodal dataset (MEG, EEG, T1, BOLD, DWI, FLASH) that is freely available to compare with other such methods (ftp://ftp.mrc-cbu.cam.ac.uk/personal/rik.henson/wakemandg_hensonrn/). Together with Hauk, we also continue to work on the M/EEG source reconstruction problem (Henson et al, 2009a; Hauk et al, 2011).
Effective connectivity – Rik Henson
We are interested in evaluating methods for inferring effective connectivity between brain regions using fMRI and MEG/EEG. Effective connectivity goes beyond functional connectivity in trying to infer directional causality, in the Granger sense of temporal precedence and/or the model comparison sense for parametrised networks. For example, we have developed a pipeline for multivariate autoregressive modelling of MEG/EEG timeseries data (Williams et al, in prep), and continue to evaluate dynamic causal models of MEG/EEG and fMRI data (Chen et al, 2009; Henson et al, 2012). We also host the Cambridge Connectome Consortium together with other groups in Cambridge (e.g, Bullmore’s group) who are interested in applying graph theoretic measures (e.g, Kitzbichler et al, 2011) and dynamical systems theory (Shriki et al, in press) to neuroimaging data.
Multivariate methods for fMRI – Niko Kriegeskorte
We use multivariate pattern-information analyses on neuroimaging data to address a wide range of questions about the content of neuronal population-code representations in different parts of the brain. In addition, we develop pattern-information methodology with a focus on fMRI, with extensions to MEG. On the one hand, we develop tools for representational similarity analysis (RSA, Kriegeskorte et al. 2008), a pattern-information technique that can be used to test computational models of brain information processing, and to relate representational geometries between brain regions, individuals, species, and measurement techniques. On the other hand, we are investigating the mechanism, potential, and limits of pattern-information fMRI, and optimal fMRI acquisition parameters. RSA and other fMRI pattern-information analyses rely on the reflection of neuronal activity patterns in fMRI hemodynamic patterns (Formisano & Kriegeskorte, 2012). High-resolution fMRI promises access to more fine-grained neuronal pattern information. However, increasing resolution in fMRI lowers the functional contrast-to-noise ratio. Moreover, it has been suggested that even standard-resolution fMRI (3mm isotropic voxels) might reveal columnar-grain neuronal pattern information. We are therefore investigating (1) whether fMRI can reveal columnar-grain neuronal information and (2) what voxel resolution and acquisition strategy (3T versus 7T, 2D versus 3D sequences) maximises the amount of neuronal pattern information reflected in the fMRI patterns.
Single cell recording – Marta Correia
While brain imaging averages together activity from millions of nerve cells or neurons, information processing takes place through detailed, high-speed communication of electrical impulses between one neuron and another. Recording such activity requires electrodes placed within the brain, sometimes in patients whose electrode grids have been implanted to help in surgical planning, but more commonly in experimental animals. A strategic collaboration with the University of Oxford allows us to carry out studies of high-level vision, attention and cognitive control in behaving monkeys, uniquely suitable because of close similarities between human and monkey brains.
Our MRI Physics research is strongly driven by the requirements of the different research programmes at the CBSU. MRI data acquisition takes place on a full-time, research-dedicated, on-site Siemens 3T TIM Trio with 12- and 32-channel headcoils.
Our areas of research include:
- Quiet MRI sequences for auditory stimulation experiments
- fMRI techniques for imaging brain areas affected by susceptibility-induced signal voids
- High-resolution and ultra-fast fMRI sequences
- Optimisation of MRI sequences for pattern classification analysis
- Development of real time fMRI applications
- Optimisation of diffusion imaging sequences for fibre tracking
MEG data analysis
We are developing and evaluating novel methods for source estimation and spatial filtering of EEG/MEG data (Olaf Hauk, Rik Henson). Recently, we started to co-register eye-movements with simultaneous EEG/MEG measurements (Olaf Hauk). We are constantly thriving to improve standard analysis pipelines (including Maxfilter software and ICA artefact correction, e.g. Jason Taylor and Olaf Hauk), and are optimising EEG/MEG recording and analysis procedures for studies on development and aging.
visit the CBU wiki farm for pages relating to