Project title: Statistical machine learning for complex system modelling: diagnosis and prediction of coronary heart diseases
Director of Studies: Dr Mu Niu
Second Supervisor: Dr Yinghui Wei
Coronary heart disease is a major cause of death both in the UK and worldwide. The change in the geometry of the heart is clinically termed remodelling, which typically occurs after a heart attack. Non-invasive tomographic imaging of cardiac shape and motion provides information about remodelling. Important diagnostic information can be obtained from the shape, deformation and stress of the heart. However, traditional clinical indices to quantify remodelling are limited to simple measures of mass and volume, discarding much of the available shape information and the dynamics of heart motion. There is a lack of comprehensive data analysis of the available image data to unpack the diagnostic and predictive information about coronary heart diseases. This project aims to make effective use of clinical image data, proposing advanced statistical machine learning methods and producing reliable inference (Niu et al., 2016; Mangion et al., 2017), to develop diagnostic and predictive tools for coronary heart diseases.
The objectives of this project are listed below:
1. A non-parametric Bayesian model will be constructed by taking the shape information as inputs and make prediction on the patient states. The important predictors for coronary heart diseases can also be identified. There will be a particular emphasis on Gaussian process and Bayesian inference in this step.
2. The heart motion can be characterised by a dynamical model. By combining the dynamical model with the non-parametric Bayesian model, we will improve the prediction accuracy of the coronary heart diseases. A novel statistical algorithm will also be developed for the parameter inference of the dynamical model.
3. The proposed methods will be applied to the existing data collected by our collaborators at University of Glasgow. We will evaluate the validity and the benefits of our proposed approaches over the conventional methods.
We will implement our developed methodology in a user-friendly software toolbox, to promote its uptake by other researchers. We will also develop clinical impact from this project by translating our research output into clinically relevant information to benefit clinicians and patients.
The student will have an excellent training experience in advanced machine learning algorithms and Bayesian statistics through this project, under the guidance of supervisors who have considerable expertise in these areas, in particular Gaussian processes. The student will have good career prospects of working as a data scientist or statistician both within and outside the academic world.
The student will also benefit from this interdisciplinary research. By working with researchers from a variety of field including statistical machine learning, medical statistics and computational biology, the student can review the recent progress in statistical modelling and applications to data science related problems, initiate new collaborations and raise new challenges.