Smart Medical Imaging

EPSRC Centre for Doctoral Training

Research

Projects

  • 2019_001 - Explaining the predictions of deep learning models in cardiology

    Dynamic imaging of the heart as it beats can provide valuable information about cardiac health. In recent years, techniques based upon deep learning have emerged that have enabled disease to be diagnosed and characterised from such images. However, like most deep learning algorithms, whilst producing highly promising results in terms of accuracy, these techniques typically lack explanatory power and often behave like a black-box.This project aims to produce an automated tool based upon deep learning that can analyse dynamic cardiac images and provide interpretable explanations as well as accurate characterisations of disease. More...

  • 2019_004 - One-step, site-specific labelling of His-tagged proteins with technetium-99m and rhenium-188

    The transition of molecular imaging agents from “serendipitous” uncharacterised small molecules with uncertain mechanisms of action to well-characterised “designer” biomolecular conjugates has not been matched by development of suitable simple chemistry for labelling with Tc-99m, which despite the advent of PET, remains the most widely used radionuclide in nuclear medicine. This project sets out to fill the gap by developing chemistry for fast, one-step radiolabelling, under mild conditions, of biomolecular targeting vectors – particularly proteins and antibody fragments such as nanobodies, scFV, minibodies etc. – with generator-produced radionuclides Tc-99m (for SPECT molecular imaging) and Re-188 (for targeted radionuclide therapy), by designing complexes that bind to the His-tag motif and optimising the synthesis of the complexes and the design of the His-tag for to achieve fast, quantitative, one-step labelling via a commercialisable kit for easy and economic use in the clinic. More...

  • 2019_005 - 3D Cardiac Magnetic Resonance Fingerprinting with Deep Learning

    In this project we aim to develop, implement and test the clinical feasibility of a novel three-dimensional (3D) Magnetic Resonance Fingerprinting (MRF) approach for free breathing multiparametric whole-heart cardiac magnetic resonance imaging. This framework will enable 3D multiparametric myocardial characterization from a single free breathing acquisition, thus providing quantitative information of multiple tissue parameters for an efficient and comprehensive assessment of cardiovascular disease. More...

  • 2019_006 - Quantitative mapping of developing fetal organs using dynamic MRI and artificial neural networks

    Quantitative T1 and T2 mapping of the fetus has potential to allow characterisation of subtle changes in tissue properties. Fetal motion presents a serious challenge to established quantitative methods and requires development of innovative motion resistant slice-selective sequences that can be reconstructed into high resolution consistent quantitative maps by combining motion correction, modelling of the dynamic responses of MR signal, and machine learning approaches to solve such complex inverse problems. The novel motion resistant quantitative techniques will be applied to fetal brain and other organs to assess their diagnostic value and facilitate comparison of fetal brain development with pre-term neonatal population. More...

  • 2019_007 - Pushing the Limits of Spatial Resolution for Whole-heart Cardiac MR Angiography: Let’s get ready for Prime Time

    We aim to develop, implement and test the clinical feasibility of a novel deep learning based undersampled reconstruction to enable free-breathing submillimetre isotropic resolution 3D whole-heart coronary magnetic resonance angiography (CMRA) and multi-contrast cardiac magnetic resonance imaging in reduced scan times. More...

  • 2019_008 - The repair of damaged tissue by mitochondrial transplantation

    This project explores the potential of transplanting mitochondria for repairing damaged tissue. While the applications for such a therapeutic approach are diverse, we will demonstrate proof-of-concept by rescuing cardiac tissue damaged by cancer chemotherapy. Not only is there an urgent need for such therapy clinically, but we have a unique rodent model of cardiotoxicity which will enable us to develop the approach and demonstrate its potential prior to exploring applications in other diseases in other organs. We will develop the basic tools for isolating and radiolabelling mitochondria to allow us to track them by nuclear and fluorescence imaging. We will study their uptake in isolated perfused rat hearts, and determine whether there is an exploitable active process involved which causes them to specifically accumulate in vulnerable tissues. Finally, we will image transplanted mitochondrial accumulation in rat hearts in vivo, and determine their capacity to repair the myocardium following chemotherapeutic cardiotoxicity. More...

  • 2019_009 - Bidentate diphosphine and dithiocarbamate chelators for radionuclide imaging with 99mTc

    This project aims to develop new dithiocarbamate and diphosphine chelators that can be attached to peptides and other small targeting molecules. The new bidentate chelator-peptide conjugates will be radiolabelled with 99mTc, enabling whole body molecular SPECT imaging of target receptors in vivo. More...

  • 2019_010 - Automated 4D Function of the Heart

    Cardiac magnetic resonance imaging is the best non-invasive method to gain information on heart performance. Recent improvements in acquisition methods enable 3D cine acquisitions in a clinical workflow, avoiding the need for multiple breath-hold acquisitions. However, there is a lack of tools available for the automatic analysis of these scans. The aim of this project is to develop new tools and pipelines for fully automated AI-enabled analysis of clinical cardiac magnetic resonance examinations from 3D+time cine acquisitions. This project will utilize the existing Siemens Frontier prototyping platform available at KCL to develop an app for automated analysis of 3D+t acquisitions using deep learning algorithms. The app will identify and delineate the left and right ventricles and the left and right atria, in every frame of the 3D cine acquisition. Machine learning methods will be applied to robustly segment the anatomy, based on transfer learning from the standard 2D+t acquisitions to the new 3D+t acquisitions. More...

  • 2019_012 - AI enabled ultrahigh field body MRI: application to bone marrow cancer imaging

    Ultrahigh field MRI (7T and above) can achieve much greater sensitivity and resolution than lower-field clinical systems, potentially benefitting a range of clinical applications including oncology imaging. However body imaging at 7T is a currently serious challenge due to the highly non-uniform radiofrequency (RF) fields produced at the 300MHz resonance frequency, meaning that parallel transmission RF (PTx) systems are needed to produce useful images. Configuration of these systems requires bespoke measurements and calculations to be performed for each new patient, creating a workflow problem that seriously limits clinical utility at present. This project will use emerging artificial intelligence (AI) methods to replace these time-consuming steps, directly relating necessary imaging parameters to readily available information such as pilot scans. The result will be transformative for 7T body MRI; we aim to develop clinical quality T1, T2 and diffusion-weighted 7T-MRI sequences, focussing on the pelvis and thoraco-lumbar spine as an exemplar. More...

  • 2019_014 - Building sensitive models of cognition using interpretable Deep Learning

    Current approaches for predicting behavioural and cognitive traits from brain imaging data are limited by a need to apply simplified models of cortical organisation to facilitate comparison between datasets. Deep Learning, on the other hand, allows direct comparisons of imaging data without the use of such prior models. It thus has the potential to significantly improve the sensitivity of neuroimaging studies as it has done for natural image analysis. Unfortunately, the mechanisms by which deep networks make predictions are not well understood making it challenging to know what features of the data are most discriminative for each predictive task. The goal of this project is to develop novel techniques for visualisation and interpretation of Deep Learning algorithms to design Deep Networks that learn biologically meaningful representations from brain imaging data, and use these features to improve understanding of the mechanisms of cognition and behaviour. More...

12345 Next