The science of ageing
Understanding neural plasticity requires understanding the
effect of age on functional systems. As one of the main aims of
the Cam-CAN project is to understand how functional networks are
affected by age, a number of ongoing analysis focuses on
developing methods for understanding functional responses.
Currently a number of methodological developments are underway
for the Cam-CAN project for analysing both incoming fMRI and MEG
We are using a multimodal approach to characterise the neural
responses to simple sensory (brief auditory tones, brief visual
checkerboard) and motor (right-hand button press) events.
Magnetoencephalography (MEG) data have a high temporal resolution
(on the order of msec), which allows us to directly assess latency
differences in these neural responses. Functional magnetic
resonance imaging (fMRI) measures have high spatial resolution (on
the order of mm), which allows us to pinpoint the location of
activity associated with a sensorimotor event.
We are developing several analysis pathways that will integrate
across the two modalities. For example, statistical maps of fMRI
activations can be used to position dipoles for MEG source
estimation. Estimates of MEG source amplitude can be used to
estimate to what extent any age-related differences in fMRI blood
oxygenation-level dependent (BOLD) contrast are due to changes in
neuro-vascular coupling rather than to bona fide differences in
neural responses. The figure shows some preliminary results from
~200 subjects in each modality, MEG (left, separately for each
event) and fMRI (for all events superimposed), consistent with the
expected pattern of activation (bilateral auditory and visual
cortex and left motor cortex).
A major challenge for current and future MEG data is to speed
the process of removing artefacts in such a large scale dataset.
If the MEG sensor net can be considered as a sphere, then
artefacts originating outside that sphere can be attenuated using
an application called MaxFilter. We are developing methods to
optimize the use of MaxFiliter to eliminate artefacts and to
employ MATLAB code to implement MaxFiltering for computationally
efficient denoising of MEG datasets. To remove artefacts
originating from inside the MEG sensor net, we used an ICA
(Independent Component Analysis)-based denoising algorithm. We
were able to select ICs (Independent Components) relating to ECG
(Electrocardiogram), VEOG (Vertical Electro-oculogram) and HEOG
(Horizontal Electro-oculogram) and remove them through a novel
bootstrap-based thresholding scheme.
Complex functional brain networks are networks of brain regions
and interregional correlations, constructed from functional brain
imaging datasets. The technological sophistication of modern
functional brain imaging, such as functional magnetic resonance
imaging (fMRI) and magnetoencephalography (MEG) allows the
acquisition of increasingly accurate and high-resolution
functional brain networks. A small number of statistical measures
provide a remarkable amount of information about the local and
global properties of these networks, as recently reviewed (Rubinov
and Sporns, 2010; Bullmore and Bassett 2011). Recent years have
seen much interest in the characterization of these networks in
the resting state, or during a no-task condition.
Functional networks in the resting state have been shown to be
heritable and reproducible, to change with administration of
neuropharmacological agents, and to change in several neurological
and psychiatric disorders, including Alzheimer's disease and
schizophrenia. In the Cam-CAN project, we aim to study the local
and global properties of complex functional brain networks in
Highly reproducible Resting-State Networks (RSNs) have been found
in fMRI data, but very little work has been done on identifying
RSNs in MEG data. Since MEG has high temporal resolution, it
allows for identifying RSNs at higher frequency bands, which would
not be visible in fMRI data. We propose to identify MEG
resting-state networks in each of the cohort of 700, and
investigate the relationship between some network features (eg.
mean degree, clustering coefficient, efficiency) and age.
To estimate RSNs, we first need to choose a measure of
connectivity. We evaluated the performance of different measures
of effective (PDC, DTF) and functional (PLI,PLV,MI, iCOH,icoh)
connectivity by comparing reproducibility of RSNs across 10
randomly chosen subjects. Our current analysis suggests that PDC
(Partial Directed Coherence) is the most robust measure of
effective connectivity and MI (Mutual Information) is the most
robust measure of functional connectivity (see figure below).
We are currently developing MATLAB code to calculate
analytically defined thresholds, to distinguish PDC values and MI
values which are statistically significant from those that are
not. After this, we will be estimating MEG resting-state networks
with PDC and MI, for different frequency bands, for each of the
cohort of 700. Then, network features for each subject will be
calculated and their relationships to age will be examined.
Finally, we will be verifying if the low-frequency MEG
resting-state networks translated to 3-D source space are related
to the fMRI resting-state networks, while the high-frequency MEG
resting-state networks are independent of the fMRI resting-state