Back to Projects

AI-enabled Imaging, Emerging Imaging

Efficient and Robust Quantitative Cardiac MRI to enable Radiomics for Heart Failure

Project ID: 2020_022

1st Supervisor: Claudia Prieto, King’s College London
2nd Supervisor: Julia Schnabel, King’s College London
Clinical Champion: Ajay Shah, King’s College London

Aim of the PhD Project:

In this proposal we aim to develop, implement and validate a novel streamlined, reproducible and fully co-registered multiparametric quantitative mapping approach from a single and efficient scan, to enable radiomic analysis of a wealth of myocardial tissue information to potentially improve heart failure diagnostic accuracy. The specific aims of the project are to:

  1. Develop a 2D cardiac Magnetic Resonance Fingerprinting (MRF) approach for simultaneous T1, T2, T2* and fat-fraction mapping.
  2. Develop deep-learning based dictionary generation and matching to enable quantification of multiple cardiac MRF parameters in computational feasible times. 
  3. Develop multiparametric radiomic analysis from the proposed 2D cardiac MRF sequence.
  4. Validate the proposed 2D cardiac MRF approach and radiomic analysis in a small cohort of patients with heart failure

Project Description / Background:

Heart failure (HF) is a raising public health issue with an estimated prevalence of >37.7 million individuals globally. HF is a shared chronic phase of several cardiac diseases and is associated with increased morbidity and mortality. Cardiovascular Magnetic Resonance (CMR) plays an important role in evaluating HF patients and potentially may provide prognostic information about disease progression and response to therapies not available with other imaging modalities. Quantitative mapping techniques have further expanded the role of CMR by quantifying several MR tissue specific parameters such as T1, T2 and T2* relaxation times. T2* relaxation time has an established role in the assessment and follow-up of iron overload cardiomyopathy and several studies have shown that changes in T1 are associated with dilated cardiomyopathy, amyloidosis, and Anderson-Fabry disease; whereas there is increasing evidence of the importance of T2 mapping in the diagnosis of myocarditis and other inflammatory conditions. Moreover, fat quantification has emerged as a valuable biomarker to characterize infarction and fatty infiltration. However, HF is often non-specific and heterogenous, and due to the variety of clinical manifestations clear-cut diagnostic criteria on CMR mapping are usually lacking.

Radiomics1 is a recently developed field that promises to increase diagnosis, prognostic and predictive accuracy for many diseases, enabling more precise personalised care which is essential in challenging syndromes such as HF. Radiomics rely on the concept that medical images contain information that reflects underlying disease-specific processes and that these relationships can be revealed via image analysis of mineable high-dimensional data. Radiomics has been mainly investigated in oncology, but a recent study has shown that radiomic analysis of T1 images provides higher diagnostic accuracy (86%) than global T1 mapping (64%) to differentiate between two cardiac conditions associated to HF2. Radiomic analysis of several CMR maps (simultaneously) should be even more powerful since different parameters provide complementary information. However, a major challenge to enable multiparametric radiomic analysis on quantitative CMR images is the lack of standardization of data acquisition and reconstruction. Multiparametric quantitative CMR acquisitions are usually performed sequentially with different MR sequences and parameters, and are not necessarily co-registered, making multiparametric analysis challenging. Moreover, quantitative mapping values are usually scanner-specific (due to several system-related confounding factors) and are not reproducible between vendors and sites, further hindering the analysis and widespread application.

Magnetic Resonance Fingerprinting (MRF)3 has been introduced to provide simultaneous MR parametric mapping from a single scan. This is achieved with an acquisition and reconstruction framework that includes three main components: 1) variable pulse sequence sensitized to parameters of interest (e.g. T1 and T2); 2) highly undersampled acquisition that introduces incoherent image artefacts; 3) dictionary-based matching for multiparametric map estimation. Cardiac MRF4 has been shown to provide simultaneous, co-registered T1 and T2 mapping in a single breath-hold acquisition, and we have recently extended this approach to also provide fat fraction (FF) information from the same scan5. Here we aim to further extend this approach to provide reproducible and fully co-registered T1, T2, T2* and FF quantitative mapping from a single streamlined and efficient cardiac MRF scan, enabling radiomics analysis of a wealth of myocardial tissue information and thus, in the future, potentially improve HF diagnostic accuracy.

This project joins expertise from MR physics, image reconstruction, machine and deep learning with cardiac clinical translation, and thus will permit the student to work and train at the interface of the different sub-disciplines. Candidates with background in physics, engineering or computer science, with a clear interest in medical imaging and clinical translation of technologies, would be suitable for this project.

References:

  1. Gillies R et al, Radiology 278(2):563-577, 2016,
  2. Neisius U et al, JACC Cardiovascular Imaging 2019, doi:10.1016/j.jcmg.2018.11.024,
  3. Ma D et al, Nature 495, 187-192, 2013,
  4. Hamilton JL et al, MRM 77(4):1446-58, 2017,
  5. Jaubert O et al, MRM 2019.

Back to Projects