Heart Podcast Can we predict coronary artery disease on CT using machine learning - insights from the SCOT-HEART trial
Dec 30, 2025
In this engaging discussion, Professor Michelle Williams, a renowned expert in cardiovascular imaging, sheds light on the groundbreaking SCOT-HEART trial. They explore how machine learning could revolutionize predicting coronary artery disease using CT scans. Michelle shares insights on high-risk plaque detection and the limitations of their findings, emphasizing the need for recalibration across healthcare systems. The conversation also touches on the upcoming SCOT-HEART 2 study, designed to assess screening benefits for asymptomatic patients, and encourages future careers in this innovative field.
AI Snips
Chapters
Transcript
Episode notes
CT Changes Management And Outcomes
- SCOT-HEART showed CT coronary angiography changes management and improves outcomes, including reduced myocardial infarction rates.
- Clinical data predicts presence of coronary artery disease but not high-risk low-attenuation plaque features.
Machine Learning Approach Used
- The team used XGBoost gradient-boosted decision trees with clinical inputs to build predictive models.
- They trained on SCOT-HEART with separate internal testing to evaluate model performance.
Clinical Data Predicts Disease But Not High-Risk Plaque
- The model predicted any coronary artery disease with AUC ~0.8 using age, sex, cholesterol, exercise test and risk score.
- The model could not predict high-risk low-attenuation plaque beyond standard risk scores.