Improving risk stratification in patients with coronary artery disease using artificial intelligence

Coronary artery disease (CAD) affects millions of people worldwide and remains a leading cause of death despite modern treatment options. Effective treatment depends on accurate risk stratification, which classifies patients based on their likelihood of adverse outcomes and helps identify those who may benefit from more intensive care. This thesis investigates how machine learning (ML) can improve risk stratification in CAD, with a focus on transthoracic echocardiography (TTE) and invasive coronary angiography.
TTE, commonly used to measure left ventricular function, provides much broader information, including details about heart valve structure and blood flow. Research involving patients with chronic coronary syndrome showed that moderate or severe valvular heart disease, especially tricuspid regurgitation, was independently associated with a higher risk of mortality, regardless of left ventricular function.
A machine learning model based on clinical and TTE data accurately predicted five-year mortality in patients with chronic coronary syndrome and performed better than traditional prediction tools. The model also showed strong performance when tested in another hospital.
Further work validated ML-based risk scores in patients with acute coronary syndrome treated with percutaneous coronary intervention. The GRACE 3.0 score predicted in-hospital mortality more effectively than the earlier version, while the PRAISE score showed limited added value.
Deep learning models developed for invasive coronary angiography achieved expert-level accuracy in identifying coronary arteries and detecting significant stenoses. These models allow for automated, quantitative assessment of disease severity.
In conclusion, machine learning shows great promise for improving cardiovascular risk assessment, but additional validation and clinical evaluation are required.