2022/06 - Precision measurement of cardiac structure and function in cardiovascular magnetic resonance using machine learning [Journal Club]-Journal Club Webinar

Video Transcription

Video Summary

This JCMR Journal Club opened with announcements about CME credit, recording availability, and audience participation via chat. The featured paper, <strong>“Precision Measurement of Cardiac Structure and Function in Cardiovascular Magnetic Resonance Using Machine Learning,”</strong> was presented by Drs. Davies and Moon.

The speakers explained that traditional human contouring in cardiac MRI is inconsistent, especially for ejection fraction, making precision more important than simple accuracy. Their team built a machine learning model using large, diverse training data from multiple institutions, scanners, disease states, and field strengths, with iterative refinement of annotations to improve robustness. The model segments endocardial and epicardial contours, tracks the valve plane, and can be deployed inline during scanning.

Results showed improved precision versus human analysis, especially for repeated scans, and potentially greater power for trials by reducing measurement noise. They also discussed prognostic findings, noting that AI-derived measures sometimes predict outcomes better than human-derived ejection fraction, and highlighted myocardial contraction fraction as a potentially superior metric.

The discussion covered challenges such as basal slice selection, respiratory motion, poor image quality, edge cases, right ventricular and congenital disease, confidence intervals, and clinical deployment. The presenters emphasized that AI should be explainable, clinically relevant, and continuously improved after deployment.

Keywords

cardiac magnetic resonance

machine learning

ejection fraction

precision measurement

contouring

myocardial contraction fraction

AI prognostic prediction

clinical deployment

basal slice selection