Joint Data Imputation and Mechanistic Modelling for Simulating Heart-Brain Interactions in Incomplete Datasets

Jaume Banus, Maxime Sermesant, Oscar Camara, Marco Lorenzi

Abstract

Cardiovascular health is a critical driver of neurological conditions, yet researchers often face a "data mismatch" where large-scale neuroimaging studies lack detailed cardiac metrics. This paper introduces a probabilistic framework that bridges this gap by combining data-driven machine learning with physics-based mechanistic modeling. By integrating a Conditional Variational Autoencoder (CVAE) with a Gaussian Process (GP) emulator, the authors can impute missing heart data in brain-focused datasets and estimate physiological parameters that are impossible to measure directly. Validated using UK Biobank data, the framework demonstrates superior accuracy over standard imputation methods and enables the generation of physically plausible "what-if" scenarios for heart-brain interactions.

Introduction

The deep connection between heart and brain health is well-documented, particularly in the context of Alzheimer’s and vascular dementia. However, a significant barrier to research is the lack of multi-modal data. Neuroimaging studies, such as the UK Biobank, typically collect extensive brain data but only basic cardiac metrics like blood pressure. Conversely, detailed cardiac studies rarely include brain scans.

This mismatch hinders the use of mechanistic models—mathematical simulations of physical processes—which require specific inputs like ventricle volumes or arterial resistance. The authors address this by developing a method to accurately impute missing cardiac features while ensuring they remain physically plausible according to cardiovascular dynamics.

Methods

The proposed joint inference framework utilizes Variational Inference to link three core components:

  1. Imputation Model (CVAE): A generative model that learns a latent representation ($z$) of a patient’s state, conditioned on available brain features ($\nu$) and clinical data. It uses this to predict missing cardiac measurements ($\hat{x}$).
  2. Mechanistic Model: A "lumped" cardiovascular model that simulates blood flow and pressure by treating anatomy as a circuit of resistances and compliances.
  3. Gaussian Process (GP) Emulator: To overcome the computational cost of mechanistic simulations, a GP is trained to "mimic" the model, mapping the latent patient state ($z$) and observed cardiac data ($x_{obs}$) directly to mechanistic parameters ($y$) like heart fiber contractility ($\sigma_0$) and peripheral resistance ($R_p$).
Probabilistic graphical model of the joint inference framework.

The study utilized a subset of 3,445 subjects from the UK Biobank who possessed both brain MRI (grey/white matter volume, WMHs) and cardiac data (stroke volume, ejection fraction, blood pressure).

Key Results and Insights

The framework's performance was evaluated against standard imputation techniques and through physiological validation.

Superior Imputation Accuracy

The model significantly outperformed standard mean and median imputation for missing cardiac features. While its performance was comparable to K-Nearest Neighbors (KNN), the joint framework provides the unique advantage of estimating internal mechanistic parameters—such as arterial compliance ($\tau$) and heart fiber stiffness ($C_1$)—which purely data-driven methods like KNN cannot provide.

Generative "What-If" Scenarios

A key strength of this generative approach is the ability to simulate heart function under varying brain health conditions. By manipulating variables like the volume of White Matter Hyper-intensities (WMHs), researchers can observe the resulting changes in simulated Pressure-Volume (PV) loops.

Simulated variation in cardiac parameters and pressure-volume loops based on brain features.

As shown in the simulations, an increase in brain lesions (WMHs) is associated with a specific cardiac "signature," including decreased stroke volume and reduced contractility. These PV loops serve as a gold standard for assessing heart function, providing a physically grounded validation of the model's predictions.

Discussion and Future Work

While the framework provides a powerful tool for multi-modal data integration, it currently relies on a simplified 1D "lumped" model of cardiovascular physics. This approximation does not account for 3D spatial details or complex cerebral blood flow. Furthermore, the model uses scalar Image-Derived Phenotypes rather than raw imaging data.

Future research aims to incorporate deep learning architectures to handle raw 3D/4D imaging data and more complex biophysical models. Such advancements could further explore the specific local vascular impairments that contribute to neurological decay, potentially identifying new targets for intervention in heart-brain pathologies.

Summary updated

Abstract

The use of mechanistic models in clinical studies is limited by the lack of multi-modal patients data representing different anatomical and physiological processes. For example, neuroimaging datasets do not provide a sufficient representation of heart features for the modeling of cardiovascular factors in brain disorders. To tackle this problem we introduce a probabilistic framework for joint cardiac data imputation and personalisation of cardiovascular mechanistic models, with application to brain studies with incomplete heart data. Our approach is based on a variational framework for the joint inference of an imputation model of cardiac information from the available features, along with a Gaussian Process emulator that can faithfully reproduce personalised cardiovascular dynamics. Experimental results on UK Biobank show that our model allows accurate imputation of missing cardiac features in datasets containing minimal heart information, e.g. systolic and diastolic blood pressures only, while jointly estimating the emulated parameters of the lumped model. This allows a novel exploration of the heart-brain joint relationship through simulation of realistic cardiac dynamics corresponding to different conditions of brain anatomy.
BibTeX
@inproceedings{banus2020joint,
  author = {Jaume Banus and Maxime Sermesant and Oscar Camara and Marco Lorenzi},
  title = {{Joint Data Imputation and Mechanistic Modelling for Simulating Heart-Brain Interactions in Incomplete Datasets}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2020},
  doi = {10.1007/978-3-030-59725-2\_46},
}
Figure