Student: Jonathan Jackson
1st supervisor: Jack Lee, King’s College London
2nd supervisor: Andy King, King’s College London
The aim of the project is to develop a non-invasive computational diagnosis pipeline for assessing the severity of coronary lesions that could easily be applied under real-time interventional settings. To achieve this, a combination of computational fluid dynamics and machine learning techniques will be used. As the full-scale CFD modelling is time-consuming, requires expertise not widely available in clinical environments, and is unsuitable for direct implementation into cathlab consoles, in this project we will seek to develop a reduced-complexity model for which the lumped shape parameters are estimated directly from the medical images using statistical learning techniques. The developed framework will be applied to 200+ clinical cases for validation, and evaluated in a pilot study in the final year of the project.
Coronary artery disease (CAD) is the most common form of cardiovascular disease, accounting for 1.8 million death in the EU, and an associated cost of €20B each year. The clinical workup of the patients with suspected coronary artery disease can be described in three stages: beginning with clinical history and physical examination, followed by a non-invasive testing, and finally invasive catherisation, designed for progressive risk-stratification and treatment. It is at the final invasive stage, that haemodyamic measurements obtained from the patient’s coronary vessels can be used to finalise the decision on whether revascularisation (as opposed to medical therapy) is an appropriate course of therapy.
Extensive research on non-invasive imaging technology has improved the detection of CAD, yielding upwards of 80% sensitivity and specificity to date. In addition, recently a CFD based strategy was proposed to estimate the haemodynamic indices directly from CT angiographic images, representing further advances in the field. However, these strategies do not apply to the patients who have been identified as having ‘high-risk’ during the initial stage of diagnosis, as these individuals proceed directly to the cathlab, bypassing the non-invasive testing. Similar trajectory applies to those suffering from an acute coronary syndrome, in whom urgent invasive catheterisation is the default course of action, excluded from the remit of the non-invasive strategies.
Further complicating the invasive diagnosis is the presence of multiple lesions and/or multivessel disease. Under these conditions, assessing the haemodynamic significance of each stenosis cannot rely on a simple fixed threshold identified from single-lesion data, and requires a quantitative framework. Currently there are limited guidelines as per how to achieve this. Determining which lesion or lesions to revascularise, necessitates a tool with which to explore varying treatment scenarios, that is able to operate from limited available data, within the permitted timeframe of the interventional settings.
In this project we propose to develop such a real-time treatment planning technology by fusing computational modelling and machine learning. A reduced scale model will be developed from the full-complexity CFD models, by characterising the geometric-haemodynamic relationship via the application of machine learning techniques. This hybrid approach has been conceptualised to retain the physical constraints of the system while exploiting the computational speed of the machine learning techniques. The new diagnostic model will be trained and extensively validated against full 3D in silico CFD simulations and invasive clinical patient data obtained from real world settings. In addition, in the final year of the project we will explore the real-time application of the new technology through a pilot clinical study.