Cross-modal autoencoder framework learns holistic representations of cardiovascular state.

Area of Science:

• Biomedical Engineering
• Cardiovascular Research
• Artificial Intelligence in Medicine

Background:

• Integrating diverse diagnostic data (e.g., cardiac MRI, ECG) for a comprehensive physiological assessment remains a significant challenge.
• Current methods often analyze modalities in isolation, limiting holistic understanding of cardiovascular health.

Purpose of the Study:

• To develop a novel cross-modal autoencoder framework for integrating cardiac magnetic resonance images (MRIs) and electrocardiograms (ECGs).
• To construct a holistic cardiovascular state representation by combining structural (MRI) and myoelectric (ECG) information.
• To demonstrate the utility of this integrated representation for phenotype prediction, MRI imputation, and unsupervised genome-wide association studies.

Main Methods:

• Development of a cross-modal autoencoder architecture designed to process and fuse information from distinct data types.
• Application of the framework to cardiac MRI and ECG datasets to learn a joint, cross-modal representation.
• Leveraging the learned representation for downstream tasks including predictive modeling, data imputation, and genetic association analysis.

Main Results:

• Successfully generated a unified representation of cardiovascular state by integrating structural MRI and myoelectric ECG data.
• Demonstrated significant improvements in phenotype prediction accuracy using the integrated representation compared to single-modality approaches.
• Showcased the capability to accurately impute complex cardiac MRIs from readily available ECG data.
• Established a novel unsupervised framework for performing genome-wide association studies on cardiovascular traits.

Conclusions:

• The developed cross-modal autoencoder effectively integrates distinct cardiovascular data modalities into a cohesive representation.
• This integrated approach enhances diagnostic capabilities, enables data imputation, and facilitates advanced genetic analyses.
• The framework offers a powerful tool for a more comprehensive characterization of cardiovascular physiology and disease.