Four-dimensional left ventricular motion clustering reveals cardiovascular phenotypes at population scale

Area of Science:

• Cardiology
• Medical Imaging
• Computational Biology

Background:

• Assessing left ventricle motion is crucial for understanding heart disease.
• Current methods often provide limited, aggregate functional data.
• Novel approaches are needed to better characterize cardiac dynamics and disease transitions.

Purpose of the Study:

• To develop and validate a novel computational framework for quantifying and visualizing left ventricle (LV) motion dynamics.
• To identify distinct LV motion phenotypes and their association with cardiovascular risk factors and outcomes.
• To enable efficient classification of patient risk and genetic predispositions using compact motion signatures.

Main Methods:

• Utilized computer vision on four-dimensional (4D) cardiac motion data from over 20,000 UK Biobank participants.
• Employed dimensionality reduction techniques on densely sampled LV point clouds to create interpretable motion signatures.
• Developed a framework to derive spatial signatures representing deviations from average cardiac motion.

Main Results:

• Identified six distinct phenogroups representing heterogeneous LV motion patterns.
• Demonstrated differential enrichment of cardiovascular outcomes and genetic risk factors across these phenogroups.
• Showcased low-dimensional motion representations as effective spatial signatures for classifying cardiac states.

Conclusions:

• Novel computational framework provides efficient representations of cardiac motion dynamics.
• Identified phenogroups offer a new classification of heart movement variations linked to disease and genetics.
• This approach enhances the ability to assess cardiovascular risk and identify underlying genetic factors from dynamic cardiac data.