Efficient and Robust Assessment of Cardiovascular Disease Using Machine Learning and Ultrasound Imaging

Lead Research Organisation: King's College London
Department Name: Imaging & Biomedical Engineering

Abstract

Heart disease is the number one killer in the world. Currently the best way of diagnosing heart disease and planning its treatment is to use a magnetic resonance imaging (MRI) scanner. However, MRI scanners are expensive and not typically used for scanning hearts in most UK hospitals. Therefore, the best diagnosis and treatment are not available to all patients. Currently the most common way of assessing heart disease is through the use of an ultrasound scanner. Although ultrasound has many advantages, it does not have such good image quality as MRI and so there are difficulties associated with its use in heart disease management. If the 'gold standard' quality of assessment from MRI could somehow be made feasible using ultrasound it would have great potential benefits for patients.

This is the aim of this project. We aim to use state-of-the-art machine learning techniques combined with rich multimodal imaging data to produce a computer model of heart disease and its associations with heart shape and motion. By incorporating MRI as well as ultrasound imaging data into the model we can exploit the power of MRI based only on ultrasound imaging. This would make possible a low cost and easy clinical pathway to the best care possible.

Planned Impact

The primary beneficiaries of the proposed research will be patients with cardiovascular diseases. The secondary beneficiaries include clinicians, in particular cardiologists, involved in the care of patients with cardiovascular diseases. In addition, the NHS and healthcare industries in general are also likely to benefit from this research. Other secondary beneficiaries include patients' families, carers and employers.

One of the primary outcomes of the project is intended to be the provision of 'gold standard' diagnosis and treatment for heart disease based purely on ultrasound data. This would widen access to high quality care and potentially result in earlier diagnosis and better patient outcomes. Therefore, in the short to mid term, patients being treated for left ventricular arrythmia will benefit from improved ultrasound-based treatment for their condition. The NHS and healthcare industries are likely to benefit from the reduced costs resulting from the use of ultrasound rather than the more expensive magnetic resonance imaging. Patients' families and carers will benefit from the improved health of their loved ones, whilst patients' employers will benefit from improved productivity arising from the better health of their workforce. In addition, the deeper insight into ventricular remodelling gained by the research performed in this project is likely to result in a longer term benefit to patients with a wider range of cardiovascular diseases, as a result of the development of new and improved treatments and diagnostic techniques.

Publications

10 25 50
publication icon
Duchateau N (2019) Machine Learning Approaches for Myocardial Motion and Deformation Analysis. in Frontiers in cardiovascular medicine

publication icon
Nascimento L (2022) The G20 emission projections to 2030 improved since the Paris Agreement, but only slightly. in Mitigation and adaptation strategies for global change

publication icon
Puyol-Antón E (2020) Automated quantification of myocardial tissue characteristics from native T1 mapping using neural networks with uncertainty-based quality-control. in Journal of cardiovascular magnetic resonance : official journal of the Society for Cardiovascular Magnetic Resonance

publication icon
Puyol-Antón E (2020) Interpretable Deep Models for Cardiac Resynchronisation Therapy Response Prediction. in Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention

 
Description Artificial intelligence (AI) techniques have been developed that enable quantification of cardiac function from both ultrasound and magnetic resonance (MR) imaging. These will greatly speed up the analysis of heart scans and make the results less variable and more reliable, hence improving diagnosis of heart disease. In addition, novel approaches have been devised to allow information from the more expensive and less available modality of MR to be used (with AI) to improve the results of ultrasound-only analysis.
Exploitation Route The next steps are to translate this work into wider patient benefit, e.g. through commercialisation. To this end, we have licensed some of the technology to a commercial partner and also produced a web-based app to make it available to academic partners for research purposes.
Sectors Healthcare

 
Description We have licensed the technology we developed to a commercial partner (Perspectum Ltd). They will use the methods as part of their COVERSCAN product, to monitor the effects of long COVID on cardiac health.
First Year Of Impact 2021
Sector Healthcare