Biophysics-based statistical learning: Application to heart and brain interactions | JoVE Visualize

Area of Science:

Cardiovascular Imaging and Physiology
Neuroimaging and Computational Biology
Biophysical Modeling

Background:

Large-scale initiatives like UK Biobank offer joint cardiac and brain imaging data.
Current research often uses association models, lacking mechanistic insights and prior physiological knowledge.
Estimating unobservable cardiovascular parameters (e.g., cardiac contractility) via biophysical models is challenging due to data limitations and ill-posed personalization.

Purpose of the Study:

To develop an approach for jointly studying heart and brain by personalizing cardiovascular models.
To constrain cardiovascular model parameter personalization using statistical relationships with brain imaging data and clinical information.
To explore the physiological plausibility and clinical relevance of these personalized models.

Main Methods:

Proposed a constrained parameter personalization approach for a lumped cardiovascular model.
Utilized statistical relationships between model parameters and brain-volumetric indices (ventricles, white matter hyperintensities) and clinical data (age, body surface area).
Validated the approach by inferring model parameters with and without specific clinical features in a large cohort (>3000 subjects) and a subgroup with atrial fibrillation (59 subjects).

Main Results:

Demonstrated that external features significantly improve cardiovascular model personalization by providing informative parameter-space constraints.
Learned physiologically plausible mechanisms through personalized cardiovascular models.
Identified significant differences associated with specific clinical conditions, such as atrial fibrillation.

Conclusions:

Constraining cardiovascular models with brain imaging and clinical data enhances personalization and reveals underlying physiological mechanisms.
This integrated approach offers deeper insights into the heart-brain relationship.
The method is applicable to large cohorts and can identify condition-specific cardiovascular differences.