Uncertainty quantification in cerebral circulation simulations focusing on the collateral flow: Surrogate model...

Area of Science:

• Biomedical Engineering
• Computational Fluid Dynamics
• Cardiovascular Physiology

Background:

• Collateral circulation in the circle of Willis (CoW) is crucial for disease mechanisms and treatment outcomes.
• Hemodynamic simulations offer a way to investigate CoW collateral circulation, reflecting systemic effects of arterial changes.
• Quantifying uncertainty in patient-specific simulations is vital for reliable results but is computationally expensive.

Purpose of the Study:

• To develop a computationally efficient method for uncertainty quantification (UQ) in CoW hemodynamic simulations.
• To create a machine learning-based surrogate model for rapid prediction of hemodynamic quantities in the CoW.
• To assess the risk of cerebral hyperperfusion (CH) by incorporating UQ into patient-specific simulations.

Main Methods:

• Constructed a machine learning surrogate model using data from 1D-0D hemodynamic simulations of the CoW.
• Implemented UQ for rapid computation of flow rate and pressure, enabling 100,000 predictions in minutes.
• Performed sensitivity analysis to identify key parameters influencing postoperative flow rate increase and CH risk.

Main Results:

• The surrogate model predicts CoW flow rate and pressure in milliseconds, drastically reducing computation time.
• UQ successfully quantified the impact of clinical data uncertainties on simulated hemodynamic quantities.
• Identified severe stenosis and small collateral artery diameter as simultaneous conditions leading to CH.

Conclusions:

• Machine learning surrogate models enable practical UQ for time-sensitive clinical applications like predicting cerebral hyperperfusion.
• Findings elucidate the relationship between collateral flow characteristics and the risk of cerebral hyperperfusion post-surgery.
• This approach enhances understanding of cerebral circulation and aids in predicting adverse events.