Introduction: Inflammatory cardiomyopathy is a significant clinical problem worldwide with diagnostic challenges and an unclear natural history. Cardiovascular magnetic resonance (CMR) is underutilised in the workup of myocarditis patients and patients instead may undergo a biopsy of the heart which carries risks. CMR underutilisation may relate in part to existing protocols requiring contrast, availability in specialist centres, expense, and a lack of CMR expertise. There is a poor prognosis in those developing heart failure or arrhythmias despite decades of research into inflammatory heart muscle diseases and a need to develop more efficient analytical approaches supporting diagnosis and risk stratification. Developing advanced cardiac imaging with artificial intelligence (AI) approaches reducing contrast, physician time and costs could address this need.
Background: The annual occurrence of inflammatory heart muscle disease has been estimated at about 1.8 million cases with the prevalence ranging from 10 to 106 per 100,000 worldwide. There were 131,376 years lived with disability and 1.26 million years of life lost attributable to myocarditis in the Global Burden of Disease Study. Inflammatory cardiomyopathy is characterised by inflammatory cell infiltration into the myocardium with a high risk of heart failure. Heterogenous aetiologies include direct injury following cardiac-trophic viral infections, as well as organ-specific or multi-system autoimmunity; regardless, immune mediated damage often mediated by infiltrating lymphocytes, is central to myocardial injury. Recent research suggests that T-cells mediate a ‘secondary’ progressive myocardial injury in some inherited myocardial disorders, consistent with the hypothesis that a dysregulated auto-reactive adaptive immune system may develop and contribute to the natural history in a variety of diseases where there is chronic myocardial injury.
CMR imaging is key to the clinical work-up of suspected or known inflammatory cardiomyopathy patients. Often patients undergo a biopsy which can miss the diagnosis with procedural complications. However, contrast enhanced CMR is a sensitive non-invasive imaging modality in inflammatory cardiomyopathy with versatile tissue characterisation including the recognition and quantification of inflammation and fibrosis. CMR is unique as it provides a virtual biopsy. If this virtual biopsy could be achieved without contrast administration in inflammatory cardiomyopathy patients, it would save time (and related cost) in the CMR scanner and avoid inserting needles in patients and exposing patients to the admittedly small likelihood of possibly severe adverse reactions to contrast agents. Machine learning (ML) methods, which are a set of AI methods that enable computers to learn tasks from data, are becoming increasingly popular in CMR.
Aims: The overarching aim of this PhD is to develop ML and AI-based approaches applied to CMR images to improve analysis and interpretation in inflammatory cardiomyopathies. This work strives towards the triple aim of healthcare delivery: improved patient experience, improved quality of care/population health whilst being more cost-effective.
Specifically, we aim in inflammatory cardiomyopathies of heterogeneous aetiology to apply ML/AI:
1) To develop automatic CMR myocardial tissue characterisation tools from manually curated ground-truth annotations to describe CMR phenotypes (presence, pattern, location and extent of myocardial inflammation or myocardial injury and fibrosis).
2) To develop automatic disease classification models incorporating above CMR tissue characterisation phenotypes, CMR structural and functional phenotypes as well as information from medical records related to medical history, physical examination, and other tests.
3) To develop tools to predict contrast enhanced CMR tissue characterisation from non-contrast CMR image information and work towards ways of reducing the need for contrast administration.
Significance and hypotheses: Firstly, we hypothesise that CMR phenotypes of inflammatory cardiomyopathy can be described with high accuracy and precision by automatic myocardial tissue characterisation by modifying existing AI tools typically developed for more general clinical scenarios. These AI pipelines will classify normal vs abnormal appearances, and the presence, pattern, location and extent of myocardial inflammation or myocardial injury and fibrosis. Secondly, using the pipelines tailored to inflammatory cardiomyopathy we hypothesise high accuracy of automatic disease classification grouping images into familiar clinical conditions such as myocarditis, sarcoidosis, and myopericarditis with or without active myocardial inflammation. Thirdly, as part of this research, we hypothesise that by developing and improving available AI tools (including our own), contrast administration for CMR can be substantially reduced with ML analyses of non-contrast scans to guide patient management. Our own pilot work (through a Barts Charity/European Regional Development Fund supported project) and those of other groups have mostly focused on small components of non-contrast CMR information and other cardiac conditions – we intend to build on this work but optimise these tools for inflammatory cardiomyopathies.
Environment and training:
This project provides a unique training opportunity to a clinical PhD student working aligned with key strengths at QMUL/Barts partnership: Barts BRC imaging/bioresource/bioinformatics cross cutting theme (Petersen/Aung), BHF Accelerator Award for inflammatory cardiomyopathies (Marelli-Berg/Mohiddin), Digital Environment Research Institute (Slabaugh/Petersen/Aung/Abdulkareem), Turing Fellows (Slabaugh/Petersen), and Barts Heart Centre which hosts the largest CMR unit and myocarditis service (Mohiddin) in the UK.
Training in machine learning will be front loaded given the clinical background of the PhD student in addition to training in CMR acquisition, interpretation, and reporting. Working alongside cardiologists and the BBR team, individuals will be recruited with CMR scans acquired, curated, and analysed. Information of the subjects will be kept on an electronic case record and will include relevant medical history, physical examination, diagnosis, test results and CMR phenotypes. These CMR parameters will be quantified and automated using supervised ML including strain, volumes, parametric mapping, and late gadolinium enhancement. The dataset of at least 400 patients will be collated that will constitute training, validation and testing sets for the supervised ML approaches to be used.
Subsequently, these quantified imaging markers will then be tested on BBR training data to create ML models which can accurately diagnose inflammatory cardiomyopathy. The above methods will be used to develop AI pipelines towards automating diagnosis of inflammatory cardiomyopathy and working towards reducing the use of contrast during CMR.
Application Web Page: https://mysis.qmul.ac.uk/urd/sits.urd/run/siw_ipp_lgn.login?process=siw_ipp_app&code1=RFQM-W6ZF-09&code2=0013