UK Biobank is a prospective epidemiological study of 500,000 participants consisting of extensive questionnaires, physical measures and biological samples, including genetics [1]. The study will link against UK National Health Service records to relate these measures to long-term health outcomes. All data from UK Biobank are available to researchers world-wide upon application for access.

The UK Biobank has begun an imaging extension that ultimately aims to image 100,000 subjects from the original cohort (now aged 45-78)[2]. Imaging includes brain, cardiac and body MRI, bone scans and carotid ultrasound. The pilot phase has scanned 6500 participants in 18 months, and produced a data release of 5000 subjects, recently made available to researchers [3].

In this talk, we overview the brain imaging component of UK Biobank, covering the imaging protocol, describing the kind of imaging data included and the kinds of biological information that can be extracted.

Imaging of 100,000 subjects requires three dedicated imaging centres operating 7 days/week with throughput of 18 subjects/day over 5 years. This daily throughput places tight timing constraints, corresponding to one subject completed every 36 minutes. UK Biobank does not target any particular disease or hypothesis, and hence the imaging protocol must be as broadly useful as possible.

Achieving this target throughput required optimization of acquisition protocols, streamlining of subject preparation and minimization of scanner dead time. The brain imaging protocol (Table 1) includes T1-weighted MPRAGE, T2-weighted FLAIR, diffusion MRI (dMRI), task-based fMRI (tfMRI), resting-state fMRI (rfMRI) and susceptibility-weighted gradient echo (SWI).

A number of protocol decisions relate to the high throughput nature of Biobank.

• Tight FOVs are efficient, but even a "conservative" 99% population compatibility would exclude 1000 participants. We conducted a study of population brain size[4] that enables our FOVs to target ≥99.9%.

• Calibration scan times accumulate. By minimizing this "dead time" (e.g. altered shim defaults) we re-gained several minutes scan time.

• Simultaneous multi-slice (multiband, MB) acquisition enabled short fMRI[5] and dMRI[6] runs without sacrificing statistical robustness or directions/b-values, respectively.

• Field map acquisitions were replaced with blip-reversed spin echo acquisition as part of dMRI.

• We hope to reduce task fMRI to 2 minutes to accommodate a short ASL perfusion scan.

Automated processing pipelines were developed, primarily based on tools from FSL, with future pipelines expanding the range of toolkits. A streamlined set of image outputs is available for each modality. Imaging-derived phenotypes (IDPs) were estimated from the processed outputs. IDPs are single values that would be appropriate for use by non-imaging experts. IDPs currently include (number of IDPs in parentheses):

• T1: volumetric estimates of tissue types and grey matter structures (33)

• dMRI: tensor- and NODDI-derived estimates in major white matter tracts (675)

• rfMRI: network-edge connectivity estimates for two network decompositions (1700)

• tfMRI: summary statistics on BOLD effect size and significance values (18)

• SWI: median T2* values in major deep grey structures (14)

The set of available IDPs is expected to expand with improved and novel functionality of processing pipelines.

The UK Biobank will comprise the largest cohort of subjects ever imaged comprehensively. The value of the resource arises not only from multi-modal/multi-organ imaging, but also from the depth of other demographic, phenotypic and exposure data. Discovery of disease risk factors will increase over time as clinical outcomes are realized in the population. For example, 5000 cohort participants are expected to develop Alzheimer's disease by 2020, rising to 10,000 by 2030.