Dissecting disease heterogeneity in cardiac patients using multimodal machine learning, modelling, and simulation method

点击翻译按钮获取中文摘要

Cardiac disease is a major cause of mortality, often leading to arrhythmic or mechanical death such as in heart failure, a condition which occurs when the heart stops being able to pump blood correctly. Despite lifestyle adjustments and life-long treatment to manage symptoms, the condition typically worsens over time and can lead to death or an urgent heart transplant. New methods are required to account for the variability of cardiac disease across a diverse patient population and to correctly evaluate severity and risks. New biomarkers derived from multimodal data could provide a stronger prognosis than current techniques by increasing diagnostic sensitivity, providing more details on the pathophysiology, predicting disease progression, and identifying the therapy option that best mitigates each individual's condition.The objective of this project is to develop novel computational methods that exploit the synergy between AI approaches and mechanistic modelling and simulation for the realisation of precision cardiology. Specific goals include identifying new multimodal biomarkers to define cardiac disease subgroups, investigating the underlying mechanisms of disease that explain these subgroups, and identifying therapeutic targets and treatments that best improve the outcomes of each subgroup. These studies will be carried out by using data-driven methods to leverage the wealth of information contained in the UK Biobank and Clinical Practice Research Datalink electronic health records, and by augmenting this information through digital twinning and simulations. Representative groups of patients will be identified by using machine learning strategies to integrate clinical data, such as cardiac magnetic resonance imaging and electrocardiogram, with patient demographics (e.g. age, sex, co-morbidities) and genetics. Unsupervised clustering strategies could provide preliminary patient subgroups. These subgroups could be compared to, or enhanced by, subgroups obtained from novel data representation strategies based on generative models (Beetz et al., 2022). This would help explain the phenotypical variability of cardiac disease in the human population and help automatically identify the biomarkers that define subgroups. Mechanistic differences between these subgroups can be explained using cardiac digital twins, which are built using an existing electromechanical pipeline that simulates cardiac activity based on the centroids of patient clusters obtained (Camps et al., 2021, Banerjee et al., 2021). In order to improve target identification in the clinic, we will investigate the efficacy of therapeutic options and their link to the new biomarkers by using simulation testing on each group's representative digital twin.In summary, the project will deliver novel strategies combining AI and simulation methods, helping to uncover new data-driven trends in cardiac biomarkers and perform cardiac simulations on a per-patient basis level to unravel mechanisms of disease. The work aims to improve current diagnosis methods and enable the discovery of more tailored treatment regimens for cardiac patients, helping to alleviate the global burden of fatal cardiac diseases.This project falls within the EPSRC research area of Healthcare Technologies, more specifically at the intersection of Clinical Technologies and Analytical Science, and addresses the Grand Challenge of optimising disease prediction, diagnosis and intervention.