Primary brain imaging documentation (this includes details of different data releases).
Acquisition: material relating to setting up and running the UK Biobank brain imaging protocol.
Analysis: material relating to the central image processing pipeline.
Genetics: Oxford Brain Imaging Genetics (BIG) interactive
web servers (with SNPxIDP summary statistic
updated server. Also, GWAS
summary statistic resources for white matter lesions and
The 3D-maps may take a few minutes to fully load. The estimation of these networks is now briefly described:
Resting-state functional MRI data from more than 4000 UK Biobank participants was combined in order to carry out a large-scale population-average mapping of the major functional networks in the brain. All subjects' structural and functional images were first spatially aligned to each other. Next, an analysis method known as "group-ICA" was used to identify the major resting-state networks/regions; this decomposes the data into a specified number of networks, and was run at two different dimensionalities (25 and 100). The dimensionality determines the number of distinct ICA components; a higher number means that the above-threshold regions within the spatial maps will be smaller.
Network matrices (representing the nodesXnodes functional connectivity) have been estimated for all subjects. Because these are symmetric about the diagonal, only the above-diagonal part of the matrices is stored by UK Biobank. See here for more information on how to work with these files.
Components that are not neuronally driven are discarded during network connectivity modelling (and so don't appear in the maps and connectomes listed above, which therefore have less than the 25/100 original number of components). The lists of good components (with component numbering starting at 1 not 0) are (for the two group-ICA dimensionalities) rfMRI_GoodComponents_d25_v1.txt and rfMRI_GoodComponents_d100_v1.txt. The full original sets of 3D spatial maps, including "bad" nodes are here: 25-components, 100-components).
The dense (voxelwise) connectome can be computed via
the MIGP group-PCA
Wishart-adjusted weighted eigenvectors file (1.1GB). Simply take
the correlation of any voxel's timeseries with all other voxels'
timeseries to get a seed-correlation map (i.e., one entry from the
full dense connectome). You can generate these single-seed correlation
maps on the fly in FSLeyes, by loading the eigenvectors file, clicking
on a seed voxel, and pressing Overlay: Seed correlation.