Rapid MRI scanning for population-level neuroimaging
Lead supervisor: Prof. Aaron T. Hess
Co-supervisors: Dr Nikolaos Priovoulos, Prof. Karla Miller
Commercial partner: Siemens Healthineers
The UK Biobank (UKB) is the largest research database of brain images in the world, having recently reached 90% of its target of 100,000 neuroimaging participants. In addition to brain scans, UKB provides a wealth of data about participants’ life history, environment, genetics, physical and cognitive attributes, and health records. Crucially, UKB participants are largely healthy when imaged and then followed long-term, providing the potential to identify early “biomarkers” that presage later ill health.
One example biomarker is the volume of the hippocampus, a brain structure that is crucial to memory that atrophies in dementia. UKB scans have been used to derive highly detailed population norms of hippocampal volume with respect to age and other demographics (Nobis 2019), against which clinical scans in new individuals can be compared to establish likelihood of early atrophy. While such biomarkers are sensitive and stable within a given scanner and protocol, they are not sufficiently quantitative to be directly comparable across scanners or protocols. In order to use these population norms in a new setting, MRI-derived biomarkers require “harmonisation” to accurately relate clinical scans to UKB-derived norms.
The purpose of this project is to develop a rapid 5-10-minute brain MRI scan that is harmonised to the UK Biobank. This rapid acquisition could be appended to the end of any clinical routine scan to assess one or more MRI-derived biomarkers. Ultimately, we would envisage the scanner producing a report for neurologists or radiologists that places the individual’s measured biomarker in the context of UKB-derived population norms. In addition to use in clinical decision making, such harmonised biomarkers would be useful for monitoring of therapeutics, drug discovery, and in clinical research studies.
We will achieve this scan time reduction through a combination of strategies including data undersampling, rapid acquisition trajectories, and a joint reconstruction that regularises the generated images, building on the large database of existing UKB images and image reconstruction expertise in Siemens and Oxford. This research will build on engineering concepts including signal compression, non-linear optimisation, and machine learning.
The project would identify a subset of imaging derived phenotypes (IDPs) most relevant to clinical populations. Using these data the minimum information required will be identified, including coarsest acceptable resolution and the image contrast, required to derive a usable IDP. Sequence undersampling and reduction strategies will be developed to achieve at least a 4-fold acceleration on top of that already used in UKB whilst maintaining image information content for IDP derivation. These undersampled acquisitions will be reconstructed using a combination of joint reconstruction (across contrasts) and deep learning, leveraging the large amount of existing image sets in the UKB (circa 100,000 subjects).
We will complement this data-driven harmonisation approach by testing alongside cutting-edge biophysical harmonisation methods under development at our centre.
Apply using course: DPhil in Clinical Neurosciences
