Precision phenotyping of inflammation with AI-powered quantitative MRI

1) Brief description of the context of the research including potential impact

Inflammation, one of the body's defense mechanisms against infection and injury, can be wrongfully directed at the body's own tissues as a result of abnormal activity of the immune system, resulting in immune-mediated inflammatory diseases (IMIDs). IMIDs afflict millions of individuals, leading to loss in productivity and increased costs to health services.

Treatments for IMIDs are now available, however these carry a high risk of serious side-effects. To decide who should receive these treatments and when they should be given (i.e. where the benefits of treatment outweigh the risks), we need to establish if inflammation is present and to monitor its activity over time. As many IMIDs affect deep tissues, clinical assessment alone is often limited, making imaging essential.

Among available imaging techniques, magnetic resonance imaging (MRI) has several advantages, including its ability to image deep tissues and to generate contrast between inflamed and non-inflamed tissue. However, conventional MRI's image contrast can only be explained as the aggregated effect of multiple tissue properties. The disentanglement of tissue properties can be achieved by quantitative MRI (qMRI), although current qMRI algorithms are prone to error, require high computation time and may not be optimised for imaging spatially heterogenous processes.

To address these limitations, we propose the development of a pipeline that incorporates novel tools for the detection of inflammation, its characterisation and its monitoring during the course of treatment. These components will enable accurate diagnosis and monitoring of inflammation, enabling doctors to customize treatment for individual patients, while also allowing for objective evaluation of newly proposed therapies in clinical trials.

2) Aims and Objectives

The aim of the project is to build a pipeline to accurately characterise inflammation, and validate it on spondyloarthritis (SpA); this 'umbrella' disease category describes a set of inflammatory diseases affecting the spine that affect young patients and carry a large burden of morbidity and reduced quality of life.

The specific objectives are to develop:
- computational modelling algorithms to analyse chemical shift encoded MRI data, that quantitatively separate the contribution of water and fat to the signal, thus creating spatial maps of biomarkers such as water/fat content. These can be used to detect abnormal changes to tissue caused by inflammation.
- a segmentation algorithm to produce metrics on the location and severity of inflammation from the biomarker maps, hence determining the inflammation phenotype.
- a registration algorithm to align MRI scans taken at different stages of the treatment process, enabling the comparison of different timepoints that allows the monitoring of changes in inflammation load in response to treatment.

These algorithms will focus on deep learning methods which possess extremely fast inference time, thus allowing their potential use in clinical care and/or trials.

3) Novelty of Research Methodology

The novelty of our research will be the development of deep learning algorithms for the detection, characterisation and monitoring of IMIDs, enabling near instantaneous inference rather than requiring the high computation time of current qMRI algorithms.

4) Alignment to EPSRC's strategies and research areas

This project aligns with EPSRC's healthcare technologies theme, specifically "Challenge two: Transforming early prediction and diagnosis", since it will build tools enabling IMID diagnosis and monitoring. This work will also allow for future development relating to "Challenge three: Discovering and accelerating the development of new interventions", as its use in clinical trials will enhance the research into new, more effective treatments.

5) Any companies or collaborators involved

None