Elizabeth Le

Title: Radiomics and Artificial Intelligence in the Prediction of Cardiovascular Events

Abstract: The impact of cardiovascular disease on society is enormous. In the UK alone, 160,000 people die from heart disease and stroke every year. Our ability to predict cardiovascular events at both the individual and population levels remains poor. This means opportunities to lower risk with drug therapy or lifestyle changes are missed.

We will study the impact of radiomics and machine learning on risk prediction in cardiovascular disease.

We have access to several well-characterised research datasets that have been acquired during previous translational vascular imaging projects. These include subjects with stable and unstable vascular disease (angina, stroke, peripheral vascular disease, aortic aneurysm, and acute coronary syndrome) as well as asymptomatic subjects with a wide range of Framingham risk scores. As well as demographics and blood work, each dataset includes imaging using different modalities (CT, PET, MRI) and tracers/contrast agents (FDG, FMISO, NaF, DOTATATE, iodinated CT contrast). The cohort of asymptomatic subjects is from a prospective study where cardiovascular events after baseline imaging have been documented (The High Risk Plaque study).

Radiomics is the extraction of quantitative data from medical images that may not be apparent to the naked eye. It has been applied in oncology to improve prediction of metastasis and death in several forms of cancer. Machine learning, as applied to medicine, has the potential to uncover links between variables (risk factors, blood results, imaging data) that might not be apparent using traditional statistical approaches such as linear and logistic regression.

We hypothesize that using these two approaches will improve the classification of patients into high and low risk groups and improve prediction of clinical events, compared to Framingham and other contemporary approaches.