1st Supervisor: Adelaide de Vecchi, King’s College London
2nd Supervisor: Pablo Lamata, King’s College London
Clinical Champion: Ronak Rajani, King’s College London
Industrial Supervisors: Valentina Lavezzo and Hernan Morales Varela, Philips Research
Aim of the PhD Project:
- To develop and improve a combined pipeline with AI-based image segmentation and shape analysis using multi-modal imaging, and computational biophysical modelling of the left and right ventricle
- To apply this tool for extracting phenotypic biomarkers to enhance post-operative risk assessment in mitral regurgitation patients
Project Description / Background:
Clinical background: In a rapidly ageing population, the prevalence of clinically significant valvular heart diseases is estimated to double by 2050, with mitral regurgitation (MR) one of the most frequent conditions. Whilst most research on MR focuses on the left ventricle (LV), right ventricular (RV) pathophysiological changes are common sequelae in these patients, with low cardiac output state secondary to RV failure a main cause of mortality following mitral valve surgery. Abrupt and complete MR eradication results in large increases in systolic wall stress in both ventricles, leading to contractile impairment. Unlike the LV, the RV is thin-walled and not designed to accommodate elevated wall stress and afterload, which result in shape remodelling. This is particularly relevant in older patients, who are likely to present pre-existing pulmonary hypertension and impaired RV mechanics that can precipitate remodelling. When present, this labels a higher-risk cohort of patients with only 30% of cases showing reversible morphological and functional changes following treatment. Therefore, understanding the effects of MR eradication on the RV holds significant clinical implications, as it potentially marks out a group of patients who demonstrate less benefit from left-sided heart valve treatment.
Project description: This project is framed in the AI-enabled decision support for diagnosis and prognosis, and will test the main hypothesis that tracking changes in RV shape and mechanics can enhance post-operative risk prediction in MR patients undergoing mitral valve replacement. Owing to its shape, the RV cannot be readily appreciated upon standard echocardiographic imaging, which is a widespread, cost-effective, and safe imaging modality. A range of AI-enabled technologies including deep learning segmentation methods, latent variable regression and subspace methods will be used to rapidly identify the shape changes and features that are a signature of elevated risk of RV failure from CT and echocardiographic imaging data. These AI identified biomarkers will be complemented with mechanistic biophysical models to propose a combined AI pipeline where morphological and functional metrics will be quantified to (a) improve preoperative planning, and (b) predict postoperative risks from both the inductive (image analysis) and deductive (biophysical models) AI reasoning, which are the two synergetic pillars of the digital twin paradigm .
Candidate background: This project will suit a candidate with a strong academic background in math, physics or engineering and an interest in programming. Knowledge of medical imaging analysis and related AI technologies, as well as basic cardiac anatomy, is desirable but not essential.
Figure 1: Schematic of the proposed pipeline, combining clinical imaging, AI-enabled image analysis, computational modelling to assess postoperative risk in MR patients.
- Corral J, …, Lamata P. The “digital twin” to enable the vision of precision cardiology. European Heart Journal, ehaa159, https://doi.org/10.1093/eurheartj/ehaa159