Uncertainty Quantification in Prospective and Predictive Patient Specific Cardiac Models

Clinical diagnosis is seldom definitive. Clinical data are noisy and sparse, and often support multiple diagnoses and potential therapies. To decide how best to treat a patient requires identifying the many possible outcomes for an individual and their corresponding probabilities. In this project we will apply the mathematics of uncertainty quantification, developed for automotive, geological and meteorological predictions, combined with biophysical models of individual patient physiology and pathophysiology to predict patient outcomes and their corresponding probabilities. This will demonstrate how patient specific computational models can be used to make prospective predictions to guide procedures and inform uncertain clinical decisions.

The use of uncertainty quantification and predictive patient specific models will be applied to patients with atrial fibrillation. Atrial fibrillation (AF) is the most common cardiac arrhythmia in the UK. In patients who do not respond to drug treatment, the pathological regions of the atria are removed or isolated through catheter ablation. However, up to 40% of patients with advanced (persistent) AF require further ablations to treat atrial tachycardia (pathological but regular activation) that develops after they have had an initial ablation to treat their AF. To reduce the number of additional procedures, this project will predict the probability that a patient will develop atrial tachycardia and the path that the atrial tachycardia will take, based on measurements recorded at the time of the initial persistent AF ablation procedure. If successful this approach would guide preventative ablations during the initial procedure to reduce the need for repeat procedures.

The UK has a long history in developing cardiac electrophysiology models and this proposed research aims to continue this trajectory, moving computational models of cardiac electrophysiology into clinical applications. This project builds on the EPSRC investment at KCL in the Medical Engineering Centre, Medical Imaging Doctoral Training Centre and Fellowships and in Sheffield in the QUINTET project and POEMS network. Development of personalised models of the heart interacts with and integrates research across the EPSRC research portfolio. Primarily this work will support and develop the UK as an international leader in the clinical translation of cardiac modelling, contributing to the clinical technologies research area (RA). The process of creating quantitative and validated models of in-vivo cardiac tissue properties will contribute towards the biomaterials and tissue engineering RA, the need to better inform model parameters and improved understanding and quantification of cardiac physiology both motivate and exploit results from the medical imaging RA and the need for improved simulations times and robustness exploits, and drives, developments in the continuum mechanics RA and high performance computing. The adoption and development of statistical parameter inference and uncertainty quantification will feed into statistics and applied probability RA. This study will support the emergent industries using models as a healthcare service to guide procedures, improving patient outcomes and reducing costs, and for creating virtual patient cohorts for designing medical devices and improving cardio-toxicity screening. This project falls within the Clinical Technologies RA and in line with EPSRC guidance is focused heavily on translation to clinical impact and engaging with clinical end users. In light of this, the primary goal of this project is:
To move computational models of the heart from deterministic models that predict a single outcome for a patient to statistical tools that predict all of the possible outcomes and their likelihoods given the underlying uncertainty and sparsity in the available clinical data.

This transformative project will bring computational biophysical models from a position of a novel deterministic analysis tool to a predictive clinical application over 4 years. At the end of the project we will have:

1) Demonstrated the ability to create statistical patient specific biophysical models of the atria to predict the outcome and guide ablation procedures. This approach will be applicable to the ventricle and across other organ systems.
2) Applied supervoxel segmentation and uncertain marching cube algorithms to generate and visualise uncertain anatomical surfaces. These ideas can be extended to interpreting and visualising segmentation uncertainty in general.
3) Shown a novel role for uncertainty quantification within the clinic, giving mathematicians new avenues to apply and translate their research and deliver impact.
4) Predicted where to optimally ablate atrial fibrillation patients to minimize atrial tachycardia following an ablation to reduce the financial and social burden of repeat ablation procedures. This will provide clinical and basic science researchers with new understanding of atrial fibrillation and how this disease responds to treatment.
5) Generated clinical pilot data necessary to underpin the application for funding for the first biophysical patient specific model-guided atrial fibrillation ablation clinical trial.
6) Created a software platform that will enable other researchers to access patient specific model creation and statistical tools and demonstrate the value of these techniques to the healthcare technology industry.
7) Created a cohort of virtual patients for performing in-silico clinical trials. This will benefit clinical research by providing a test bed for new treatments and for device companies aiming to evaluate new technologies on virtual patients.