Assessment of valvular function in over 47,000 people using deep learning-based flow measurements

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Valvular heart disease presents a significant global health challenge.
• Understanding normal valvular function variation is crucial for early disease detection.

Purpose of the Study:

• To develop and apply a deep learning model for analyzing velocity-encoded MRI data.
• To establish sex-stratified reference ranges for valvular function phenotypes.
• To identify associations between cardiovascular risk factors and valvular function.

Main Methods:

• Utilized deep learning on velocity-encoded MRI from 47,223 UK Biobank participants.
• Calculated eight valvular and aortic phenotypes.
• Established reference ranges in 31,909 healthy individuals.

Main Results:

• Identified an annual aortic valve area decrement of 0.03cm² in healthy individuals.
• Validated associations between phenotypes and conditions like mitral valve prolapse and aortic stenosis.
• Linked higher ApoB, triglycerides, Lp(a), and glycoprotein acetyls to increased aortic valve gradients.
• Velocity-derived phenotypes predicted future valve surgery risk.

Conclusions:

• Machine learning quantification of UK Biobank data provides the largest assessment of valvular function in the general population.
• Velocity-derived phenotypes serve as early risk markers for valvular disease and surgery.
• This approach enhances understanding of cardiovascular disease risk factors impacting valvular health.