1st Supervisor: Steven Niederer, King’s College London
2nd Supervisor: Alistair Young, King’s College London
Clinical Champion: Aldo Rinaldi, King’s College London
Aim of the PhD Project:
Imaging cardiac function, form and physiology through time
- Automatically quantify shape and function with uncertainty form cardiac images.
- Implement a Bayesian update for images to integrate measurements over multiple scans
- Learn the mapping from Form/Function to Physiology to track biomarkers through time.
Project Description / Background:
Precision Medicine in Cardiology (Precision Cardiology) is the emergent strategy for cardiovascular disease treatment and prevention that accounts for individual variability. A digital twin is a digital replica of a specific physical entity. It encodes physics, physiology, population data and patient specific measurements and combines these with AI techniques for updating and interrogating the model. Healthcare is the industry most likely to be disrupted by the digital twin technology, as it enables building continuously-adjustable patient-specific models based on tracked health and lifestyle data that can anticipate a patient needs. The aim of this PhD will be to develop and test data science, modelling and simulation methods for creating a cardiac digital twin that learns from clinical data and images over time.
To realise the cardiac digital twin requires developing methods for continually updating a digital representation of a patient’s cardiac anatomy, function and physiology through time. This shifts data analysis from considering a single snapshot of a patient in a single scanning session to analysing how measurements evolve as a disease progresses or in response to a therapy. This allows continuous digital twin learning, prediction and validation over time.
To realise the digital twin will require robustly analysing images and integrating repeat clinical measurements from a patient though time into a single digital representation of a patient. This PhD will develop and test solutions for four elements of the digital twin:
- Conventional cardiac clinical analysis uses biomarkers from clinical scans to inform decisions. Patients may be scanned multiple times during their diagnosis and following treatment, but each scan is interrogated without using prior knowledge from previous scans. This project will test the hypothesis that scan analysis can be improved by using prior scans to analyse subsequent images.
- In current clinical practice a scan is often classified as diagnostic or not diagnostic. The quality of the scan is not represented in uncertainty in derived indexes due to image quality. If prior scans are used to interpret subsequent scans the quality of the scan needs to be encoded in derived indexes to avoid errors propagating across multiple scans. This project will test the hypothesis that scan quality can be encoded in uncertainty in image derived indexes.
- The first image of a patient could be used to initialise the digital twin. However, this will increase the impact of any errors in the first scan on subsequent scans. This project will test if priors of form and function can be derived from patient demographics to provide the best estimate of the patient’s heart prior to any scan being performed. This can then be updated with subsequent scans.
- Cardiac shape and function are linked through cardiac physiology. Continuous monitoring of shape and function can be used to infer indexes of cardiac physiology (passive stiffness, activate contraction or conductivity). We will test if patient form, function, demographics or physiology are most predictive of patient outcome.
Developing and demonstrating a theoretical and practical framework for continuously integrating data into a predictive digital representation of a patient will provide a step change in patient monitoring, increased precision in clinical measurements of form and function, encoding of image quality in indexes and a mapping for conventional form and function from images onto patient physiology.
The project would be suited to a computer science, engineering or applied mathematics student familiar with python, machine learning libraries, image processing, parameter inference, Bayesian statistics, and computational science. Experience using high performance computing or GPU clusters would be beneficial.