Magnetic resonance imaging (MRI) has transformed the way we look through the human body by offering exquisite soft-tissue contrast in high-resolution images, noninvasively. This has made MRI the gold-standard imaging technique for diagnosis and monitoring of many diseases. However, conventional MRI scans do not produce "quantitative" measurements, i.e. standardised measures, and therefore it is difficult to compare MRI images acquired at different hospitals, or at different points in time, limiting the potential of this imaging technology for advanced diagnostic and monitoring precision. Quantitative MRI (qMRI) aims to overcome this problem by yielding reproducible measurements that quantify tissue bio-properties, independent of the scanner and scanning times. This could transform the existing scanners from picture-taking machines to scientific measuring instruments, enabling objective comparisons across clinical sites, individuals and different time-points. But unfortunately qMRIs have excessively long acquisition times which currently create a major obstacle for their wide adoption in clinical routines. Therefore, the main goal of this project is to develop new computational methodologies based on compressed sampling and machine learning that will substantially reduce the scan times of qMRI. Compressed sampling techniques enable efficient acquisition of signals and images from tightly constrained sensor/imaging systems. They have been recently applied to address the issue of scan time in qMRI, but these techniques require much better computational methods for removing image compression artefacts at higher acceleration (compression) rates needed for this application. The project aims to address this gap through advanced machine learning-based models and appropriately chosen datasets to train them. The research has two streams of beneficiaries: (i) A large community of UK and international clinical academics that use qMRI techniques for their research on precision imaging and evaluation of diseases such as cancer, cardiac or neurodegenerative disorders, each with significant socioeconomic impact. The outcomes of this project would allow these studies to become more available and more economically feasible. (ii) A large community of UK and international non-clinical academics/professionals who work on compressed sampling inverse problem techniques, motivated by variety of other sensing/imaging applications that could benefit in their studies from methodologies developed by this project. A number of activities have been carefully designed to effectively engage with beneficiaries of this research. These activities include co-production and validation of knowledge with clinical academics and healthcare industry as our project partners, publishing of the results in leading academic journals/conferences, a project website to publicize up-to-date project advances and share open-source software and demonstrators, and a workshop with field specialists and national academic and non-academic stakeholders in medical technologies.

4D flow Magnetic Resonance Imaging (flow-MRI) is a non-invasive flow imaging technique widely used in medicine and engineering to measure velocity fields in three spatial and one time dimension (4D). For example, it is used to measure the velocity of blood in the heart and surrounding vessels to identify anomalies such as aneurysms and stenoses. The velocity measurements become increasingly noisy, however, as the spatial resolution is increased. To achieve acceptable signal-to-noise ratio (SNR), scans are often repeated and averaged, leading to long acquisition times. This proposal is to extend the capabilities of flow-MRI using Bayesian physics-constrained algorithms that automatically generate the most likely digital twin of a flow from noisy and sparse 4D flow-MRI data. These methods increase the accuracy and the spatiotemporal resolution of flow-MRI by 10 to 100 times, provide quantitative estimates of derived flow quantities that are difficult to measure, and enable the imaging of flows whose short length and/or time scales cannot be captured using state-of-the-art flow-MRI techniques. In porous media flows, for example, these methods will provide velocity fields, stress tensors, and derived quantities far beyond the accuracy of current state-of-the-art flow-MRI, leading to better understanding and new discoveries. In medical imaging, these methods will also enable patient-specific modelling and, if successful, will lead to increased adoption of 4D flow-MRI by clinicians. This would reduce patient scan times, replace invasive techniques such as cardiac catheterization, and permit the imaging of smaller vessels such as those found in neonatal and fetal cardiology. In my PhD I developed these methods and showed that, to obtain a given accuracy in axisymmetric and 2D planar flows, they reduce the required flow-MRI data by 10 to 100 times. The aim of this fellowship is to extend the methods I developed during my PhD from 2D and 3D steady flows in rigid geometries to 4D flows in flexible geometries, to scope out challenges posed by in-vivo cardiovascular haemodynamics, and to disseminate these methods widely. In this proposal I will focus on flow-MRI, but note that these methods could be extended to other velocimetry methods such as PIV.