Imaging Data

#CheadleDA UK Biobank Data Analysts
#CheadleDA UK Biobank Data Analysts The helpers that keep the community running smoothly. UKB Community team Data Analyst
  • Updated

Introduction 

UK Biobank performs a variety of imaging scans on participants and provides the results to researchers in multiple formats, including DICOM, NIFTI, and MAT files. For a brief overview, UK Biobank holds Magnetic Resonance Imaging (MRI) data including scans of the brain, abdomen and heart. We also provide dual energy x-ray absorptiometry (DXA) data of the body, ultrasound scanning data of the carotid artery, as well as Optical Coherence Tomography (OCT) of the retina. This article will describe the different types of imaging scans, as well as direct you to where to find the new, regularly updated tranches of imaging data on Showcase.

Cardiac Magnetic Resonance Imaging

Multiple types of MR sequences are used to target the heart. The resultant images are categorised into aortic distensibility images, left ventricular outflow tract images, cine tagging images, long axis heart images, short axis heart images, shMOLLI (definition) images, and blood flow images. We also provide the scout images for completeness. These image types have a wide variety of uses, including (but not limited to) providing data on cardiac, aortic and left ventricular size and function. All files are available in DICOM format. 

Image derived phenotypes (IDPs) are also available for heart MRI. IDPs are quantitative and interpretive indicators of imaging data, whereby scanner images do not need to be directly processed. The IDPs for CMR were made via an automated machine-learning-based analysis pipeline, which was developed by Bai et al. (2020), and provide cardiac and aortic structural and functional data. More information about IDPs for heart MRI can be found in data-field 25931 on Showcase. 

Cardiac MRI was conducted using a 1.5 Tesla (Siemens Healthcare, Germany) scanner, using electrocardiogram (ECG) processing for cardiac synchronisation. The protocol lasted 20 minutes in length, which is a part of a 30-minute combined cardiac MRI and abdominal MRI protocol. More information about the particulars of cardiac MRI protocols and measures can be found within category 102 in Showcase by navigating through subcategories and checking the resources sections. Some documents of note are resource 349 which provide details of the CMR procedure, as well as resource 15146 which gives information on the cardiac MRI exam rationale and protocol concepts.

Abdominal MRI

The abdominal MRI were acquired on the same 1.5T scanner following acquisition of the cardiac MRI. These scans provide a single image volume covering the body from neck-to-knee and multiple sequences with focussed fields of view on each of the kidney, liver, and pancreas. Scans provide data on liver fat percentage and iron, liver inflammation factor (LIF), abdominal subcutaneous adipose tissue (ASAT), visceral adipose tissue (VAT) and posterior and anterior thigh muscle mass (right and left). All abdominal MR image files are provided in DICOM format. Further information detailing the procedures used to collect abdominal data can be found on the showcase category 105 by navigating through the subcategories and resources sections. For example, refer to resource 348 for information on abdominal MRI procedures and resource 2005 for the abdominal scan protocol. 

IDPs are also available for over 40,000 participants with abdominal MRI scans which provide data on organ composition, such as liver, kidney and lung volume. They were developed using multiple versions of Calico's machine learning IDP pipeline (Basty et al. 2022; Liu et al. 2021; Whitcher et al. 2022). For more information on the IDPs available for abdominal MRI, please see data-field 21094 on Showcase and refer to the published papers by the above authors. 

Brain MRI

UK Biobank brain MRI measures offer structural, functional and diffusion views of the brain. This includes resting functional MRI (fMRI), T1-weighted structural imaging, T2 FLAIR imaging, arterial spin labelling, diffusion, and susceptibility-weighted imaging. Additionally, a task fMRI was also conducted which involves participants completing a faces and shapes emotion task whilst in the scanner. All data was collected using a 3T (Siemens Skyra) scanner across a 35-minute imaging protocol, and is produced in zipped DICOM file format, found under Showcase category 507. These DICOM files are also converted into sets of pre-processed, intermediate and fully processed NIFTI format files for researchers to use, which can be found under category 508. It should be noted that NIFTI T1 and T2 sequence data are defaced to further protect participant anonymity. Access to the T1 and T2 DICOM data is restricted to only those with a valid scientific justification for not using the defaced images. If you wish to make a request for this data, contact the UKB Access team by email or by using the community forum to submit a ticket to provide your justification.

IDPs are available for brain MRI in various formats, including task and resting fMRI, T1-weighted, T2 weighted, arterial spin labelling, diffusion, susceptibility-weighted imaging, and surface-based analysis of resting and task fMRI. These IDPs were derived by processing the MR DICOM files using the UK Biobank brain imaging pipeline, which is an automated image processing and quality control pipeline using software packages like FSL and Freesurfer. You can find more information about the pipeline at resource 2006 on Showcase, as well as further details on the Brain Imaging Online Resources page, and a full description of the core pipeline in a paper by Alfaro-Almagro et al. (2018).

Connectome data is also available to researchers under category 509 in Showcase, offering both functional and structural connectivity data. Functional connectivity data is available under category 203 which describes inter-regional interactions in brain activity, whilst structural connectivity is available under category 204 which provides data on white matter axonal architecture. This dataset was generated by Mansour et al. (2023), and for further details about how this data was generated you may refer to their research paper.

More information about the specifics of brain MRI procedures is available within category 100 in Showcase by navigating through the subcategories and resources of interest. Particularly, resource 1977 and resource 2367 may be of note which detail imaging acquisition and the scanning protocol, respectively.

Whole-Body DXA

Following the MRI scanning sessions, participants will undergo a whole body DXA imaging scan, which is conducted using an iDXA instrument (GE-Lunar, Madison, WI). This provides data on several different types of images, including a whole-body image, as well as hip, knee, and lateral thoraco-lumbar spine. There are multiple output types available, including some files extracted to the DICOM format, and numerical measures of bone area, mineral content and density, together with lean and fat mass measures. Further information detailing DXA imaging can be found on the DXA assessment page on Showcase (category 103) by navigating through the resources and subcategories sections. For example, more information about the raw DXA images can be found in data-field 20158, and resource 502 describes the scanning procedure. 

There are IDPs available from the whole body DXA scanning sessions, including body composition results outputted by the DXA system itself, as well as bone composition results derived from the AUGMENT study, which stands for the AUtomated Generation of Musculoskeletal phENotypes from the UK biobank exTended imaging study. For more information about body composition IDPs, please refer to category 124 on Showcase. For more information about bone composition IDPs, please refer to category 125 on Showcase. 

Carotid Ultrasound 

The carotid ultrasound measures are taken using a CardioHealth Station device which provides imaging data on the carotid arteries. The ultrasound data within UK Biobank is also provided in multiple formats, including numerical measures of the carotid intima-media thickness (CIMT), images of the carotid artery (in a jpeg format) and carotid sweep scans in .cine format. The cine scans can be found either in their raw .cine format, or in a .mat format. To find further information on the carotid ultrasound measures, navigate to the Showcase category 101. More specifically, resource 511 provides useful information on the ultrasound procedure. 

Optical-coherence tomography

Optical Coherence Tomography (OCT) is used to produce cross-sectional images of the retina. At UK Biobank, measures are taken using a TOPCON 3D OCT 1000 Mk2 (Topcon Corporation, Japan) which provides a 3D scan of the retina as well as a magnified photograph of the fundus. Further information about the types of OCT data available can be found in category 10016 on Showcase, particularly within resource 100237 which provides further information on the scanning procedure. 

There are IDPs available for OCT measures, such as thickness of the macular, retinal nerve fibre layer, inner nuclear layer and many more. These variables have been derived from various external researchers who have returned their data to UK Biobank. More can be found out about these measures by navigating to category 100079 on Showcase and checking the notes and resources sections for each data-field.

Instances 

At UK Biobank, the term “instance” refers to which visit the participant has made to the assessment centre. At different instances, unique data collection procedures are followed, in which certain sets of assessments are made on participants. For more detailed information on instances, please refer to the article What is an instance index. This is important to note since most imaging data, excluding OCT, was not collected during the baseline visit (instance 0) or during the repeat assessment (instance 1). The main imaging study only began during instance 2, where some participants were invited to return to the UKB assessment centres for imaging scans.  Instance 2 refers to the first imaging visit assessment, which began being collected in 2014, whilst instance 3 refers to a repeat imaging assessment of these participants starting from 2019. Instance 2 and instance 3 are therefore currently both active, whereby new participants to UKB are still being recorded under instance 2 and previously scanned participants are still returning for an instance 3 visit.

Mismatch of data between Showcase and the UKB RAP 

Occasionally, you may notice a difference between the amount of data that is represented in Showcase, versus when you begin to analyse your data on the UKB RAP. This can frequently happen when wanting to work with derived data values from imaging scans, as opposed to the raw data. This also includes when you may want to work with converted file formats, as mentioned for NIFTI and DICOM formats in the Brain MRI section above. In many cases, this delay is because derived data is provided to UK Biobank by external researchers, rather than being generated within UK Biobank itself. Instead, the data we provide will usually only consist of raw scanning images. For further reasons why this discrepancy can happen, please refer to the article Why is data missing from my RAP project.

Analysis Tools

There are a range of different ways that analysis can be undertaken for imaging data. Individuals less comfortable with image analysis tools may be interested in using the IDPs to begin analysing data without the need for extensive image processing.  For a more hands-on approach at image analysis, open-source tools such as FSL and FreeSurfer can be downloaded in the UKB RAP. You can find FSL and FreeSurfer pre-downloaded on the IMAGE_PREPROCESSING feature dropdown menu on the JupyterLab app, as shown in the screenshot below.

 

img_analysis_rap.JPG

There are also multiple visualisation options for both noninteractive and interactive viewing including python packages like pydicom and nibabel (which allow for non-interactive image viewing), and Mango papaya (which allows interactive viewing). Individuals interested in machine learning frameworks may wish to use options like TensorFlow, Pytorch and keras within Jupyterlab. For more information on using JupyterLab in the UKB RAP, please see the UKB-RAP documentation. For more information about image analysis on the UKB-RAP, please refer to this video produced by DNAnexus on image analysis and visualisation methods. Efforts are ongoing to improve the availability of resources and therefore this list may not be exhaustive. Further updates will be publicised here, on the UK Biobank GitHub account, and on the DNAnexus tool library.

 

 

References

  • Alfaro-Almagro, F., Jenkinson, M., Bangerter, N. K., et al. (2018). Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank. NeuroImage, 166, 400–424. https://doi.org/10.1016/j.neuroimage.2017.10.034
  • Bai, W., Suzuki, H., Huang, J., et al. (2020). A population-based phenome-wide association study of cardiac and aortic structure and function. Nature Medicine, 26(10), 1654–1662. https://doi.org/10.1038/s41591-020-1009-y
  • Basty, N., Sorokin, E. P., Thanaj, M., et al. (2022). Abdominal Imaging Associates Body Composition with COVID-19 Severity. medRxiv. https://doi.org/10.1101/2022.02.22.22270091
  • Liu, Y., Basty, N., Whitcher, B., et al. (2021). Genetic architecture of 11 organ traits derived from abdominal MRI using deep learning. eLife, 10, e65554. https://doi.org/10.7554/eLife.65554
  • Whitcher, B., Thanaj, M., Cule, et al. (2022). Precision MRI phenotyping enables detection of small changes in body composition for longitudinal cohorts. Scientific Reports, 12(1), 3748. https://doi.org/10.1038/s41598-022-07556-y

Related to

Was this article helpful?

0 out of 1 found this helpful

Have more questions? Submit a request

Comments

0 comments

Article is closed for comments.