FEASIBILITY OF VASCULAR REMODELING PARAMETER ESTIMATION FOR ASSESSING HYPERTENSIVE PREGNANCY DISORDERS

Georgios Kissas, Eileen Hwuang, Elizabeth Thompson, Nadav Schwartz, John A. Detre, Walter R. Witschey, Paris PerdikarisbioRxiv 2021 · DOI

Abstract

Hypertensive pregnancy disorders (HPDs), such as preeclampsia, are leading sources of maternal and fetal morbidity. While non-invasive imaging—including ultrasound and magnetic resonance imaging (MRI)—is critical for monitoring high-risk pregnancies, current clinical interpretations rely on ad-hoc standards like pulsatility and resistivity indices. These metrics offer limited insight into the physical mechanisms driving disease. This research presents a patient-specific computational framework that integrates Bayesian inference with reduced-order fluid dynamics to infer hidden vascular remodeling parameters, such as resistance and compliance. By utilizing sparse and noisy MRI measurements, the model can predict absolute pressure and velocity, successfully distinguishing between normotensive pregnancies and those destined to develop hypertension.

Introduction

Hypertensive pregnancy disorders affect approximately 5% of pregnancies in the United States. Despite their prevalence and danger, the underlying physical causes of these conditions remain poorly understood. Clinicians currently rely on Doppler ultrasound or MRI to measure blood velocity in the uterine arteries, calculating indices that serve as proxies for vascular health. However, these indices fail to provide a direct physical interpretation of vascular remodeling. Furthermore, the spiral arteries—the suspected origin point of preeclampsia—are too small to be resolved by current imaging technology. There is a clear need for a methodology that bridges the gap between raw imaging data and meaningful physiological assessment.

Methodology: The Digital Twin Approach

The proposed framework treats blood flow as a physics-constrained problem rather than a simple image-processing task. The system consists of two primary components:

  1. The Forward Model: A reduced-order (1D) version of the Navier-Stokes equations simulates pulsatile blood flow through a network of compliant vessels. To account for the microvasculature at the network's outlets, the researchers utilized 0D "Windkessel" models to represent downstream resistance and elasticity.
  2. The Inference Engine: Clinical MRI data is notoriously sparse and noisy. To overcome this, the framework employs Bayesian inference (specifically Slice Sampling). The model takes a patient’s specific arterial geometry and velocity measurements, then iteratively estimates vascular parameters—resistance ($R$), compliance ($C$), and equilibrium cross-sectional area ($A_0$)—until the simulation aligns with the observed data.

This approach allows for the estimation of "hidden" variables that cannot be measured directly, most notably the absolute blood pressure within internal arterial networks.

Experimental Validation

The framework was validated using three distinct datasets to test its robustness across different physiological conditions:

  • Aorta/Carotid Baseline: A 27-year-old healthy female volunteer provided data to test the model on both synthetic and real MRI measurements.
  • Normal Pregnancy Subject: A volunteer at 18.7 weeks of gestation who maintained a healthy pregnancy.
  • Pre-HPD Subject: A volunteer at 19.1 weeks of gestation who appeared healthy during the initial scan but was diagnosed with preeclampsia four months later.

Key Results and Insights

The study demonstrated that physics-informed inference can extract high-fidelity data from clinical settings where traditional machine learning might struggle due to small sample sizes.

Accuracy and Reconstruction

In the aorta validation experiments, the model reconstructed flow rates with less than 1% error for synthetic data and between 3-6% error for real-world MRI measurements. This confirms the model's ability to accurately interpret complex, pulsatile flow.

Predictive Capabilities

The most significant finding was the model's ability to distinguish between the normal and pre-HPD subjects months before a clinical diagnosis. The pre-HPD subject exhibited a significantly higher predicted systolic/diastolic pressure in the uterine arteries (152/65 mmHg) compared to the normal subject (108/33 mmHg).

Vascular Signatures

The model identified specific vascular signatures associated with HPD. The pre-HPD subject showed higher vascular resistance and lower compliance in the external iliac arteries. This supports the physiological theory that HPD involves a failure of the maternal vessels to properly "remodel" or soften to accommodate the increased blood flow required by pregnancy.

Discussion and Future Work

By providing a probability distribution rather than a single deterministic value, the Bayesian approach quantifies uncertainty, a critical requirement for clinical decision-making. However, the authors note several limitations, including the small cohort size and the simplifications inherent in 1D modeling, which may ignore smaller arterial branches.

Future research will focus on scaling this methodology to larger patient groups and integrating more complex 3D fluid dynamics as imaging resolution improves. Ultimately, this "digital twin" framework represents a significant step toward a non-invasive, early-detection screening tool for preeclampsia, potentially allowing for earlier intervention and improved maternal-fetal outcomes.

Summary updated

Abstract

ABSTRACT Hypertensive pregnancy disorders, such as preeclampsia, are leading sources of both maternal and fetal morbidity in pregnancy. Non-invasive imaging, such as ultrasound and magnetic resonance imaging (MRI), is an important tool in predicting and monitoring these high risk pregnancies. While imaging can measure hemodynamic parameters, such as uterine artery pulsatility and resistivity indices, the interpretation of such metrics for disease assessment rely on ad-hoc standards, which provide limited insight to the physical mechanisms underlying the emergence of hypertensive pregnancy disorders. To provide meaningful interpretation of measured hemodynamic data in patients, advances in computational fluid dynamics can be brought to bear. In this work, we develop a patient-specific computational framework that combines Bayesian inference with a reduced-order fluid dynamics model to infer remodeling parameters, such as vascular resistance, compliance and vessel cross-sectional area, known to be related to the development of hypertension. The proposed framework enables the prediction of hemodynamic quantities of interest, such as pressure and velocity, directly from sparse and noisy MRI measurements. We illustrate the effectiveness of this approach in two systemic arterial network geometries: an aorta with carotid and a maternal pelvic arterial network. For both cases, the model can reconstruct the provided measurements and infer parameters of interest. In the case of the maternal pelvic arteries, the model can make a distinction between the pregnancies destined to develop hypertension and those that remain normotensive, expressed through the value range of the predicted absolute pressure.
BibTeX
@misc{kissas2021feasibility,
  author = {Georgios Kissas and Eileen Hwuang and Elizabeth Thompson and Nadav Schwartz and John A. Detre and Walter R. Witschey and Paris Perdikaris},
  title = {{FEASIBILITY OF VASCULAR REMODELING PARAMETER ESTIMATION FOR ASSESSING HYPERTENSIVE PREGNANCY DISORDERS}},
  year = {2021},
  doi = {10.1101/2021.11.10.468132},
  howpublished = {bioRxiv},
}
Figure