Back to Projects

Emerging Imaging, Affordable Imaging

AI-driven machine independent quantitative MRI

Project ID: 2023_026

1st Supervisor: Dr Shaihan Malik, King’s College London
2nd Supervisor: Dr Jorge Cardoso, King’s College London
3rd Supervisor: Prof Xavier Golay, UCL and Gold Standard Phantoms
Additional External Advisor: Prof Scott Swanson, University of Michigan
Clinical Supervisor: Prof Alexander Hammers, King’s College London


Project description/background:

Magnetic Resonance Imaging (MRI) is widely used for imaging soft tissues throughout the body for investigating a huge range of different diseases/injuries.  Currently, the standard practice is to use ‘qualitative’ images –  i.e. the intensity of different tissue types do not have a special significance; instead, images are interpreted subjectively by an expert radiologist.  To mitigate this, there is now a growing body of research and development aiming to create ‘quantitative’ MRI (qMRI) scans, where the signal value in each pixel is an actually meaningful measurement of the biological tissue, thus making image interpretation more objective and diagnosis more accurate. Quantitative imaging has a tremendous potential; from more precise diagnosis, to the ability to compare subjects between themselves and across time, and to help ‘big data’ analysis, leading to a better understanding of disease.


The Problem:

Although quantitative MRI is starting to be used more widely, the results obtained are neither truly quantitative nor generally reproducible: they change depending on how the measurement is made, and on which type of scanner was used. The aim of this research is to develop an objective quantification of human tissue – i.e. quantitative measures which can be made robustly across a wide range of MRI scanners. To do this we will need to switch focus; instead of measuring the properties of water in human tissue, we aim to accurately quantify properties of the matrix of proteins and large molecules (referred to as the ‘semisolid’) which is a key constituent of the tissue. We believe that if these properties can be accurately measured, they would be likely to be less variable between scanners. It is currently very difficult to quantify these properties accurately and efficiently, so this project will couple advanced artificial intelligence models (e.g. recurrent/transformer networks) with cutting edge scanner technology to change how this is done.


This project:

We aim to establish measurements that can obtain equivalent results both with expensive MRI scanners used within hospitals today, and lower cost alternatives that are now emerging. This will build towards the longer-term aim of improving the accessibility to MRI technology.

The project will involve a fusion of AI and physics, using both computational and experimental work. It would be well suited to a candidate with background in computer science, physics or (biomedical) engineering and will be an excellent training opportunity for someone interested in either a career in academic biomedical engineering research or technical development within industry. It will also provide strong training fundamentals for a career in advanced AI and data science.


The project will have two major parts:

  1. Developing AI methods for predicting MRI signal responses from complex tissue models, and for inferring tissue parameters from noisy MRI data.
  2. Working on a range of cutting-edge MRI scanners operating at a range of magnetic field strengths (64mT to 7T), all available at the St. Thomas’ campus, to implement novel methods driven by the AI models.


The student will also have the opportunity to work with UK based company Gold Standard Phantoms, to develop and characterise synthetic materials needed for benchmarking these quantitative methods.


Back to Projects