Reconstructing blood flow in data-poor regimes: a vasculature network kernel for Gaussian process regression.

Area of Science:

• Biomedical Engineering
• Computational Fluid Dynamics
• Medical Imaging

Background:

• Accurate blood flow reconstruction is vital for clinical applications.
• Limited clinical data, such as from transcranial Doppler ultrasound, hinders machine learning model training.
• Existing methods struggle with data-poor regimes in vascular modeling.

Purpose of the Study:

• To develop a Gaussian process regression approach for near-real-time blood flow reconstruction in data-poor vascular networks.
• To introduce a novel kernel reconstruction methodology that incorporates spatiotemporal and vessel-to-vessel correlations.
• To ensure predictions satisfy the conservation of mass principle.

Main Methods:

• Utilizing Gaussian process regression with empirical kernels derived from physics-based simulations.
• Developing a novel kernel reconstruction methodology for vascular networks.
• Employing stochastic one-dimensional blood flow simulations to capture uncertainties.

Main Results:

• The proposed method enables blood flow reconstruction in vessels lacking direct measurements.
• All predictions made using the proposed kernel satisfy the conservation of mass.
• The model demonstrated effective performance on diverse vascular geometries, including bifurcations and the circle of Willis.

Conclusions:

• The developed Gaussian process regression approach effectively reconstructs blood flow in data-poor vascular networks.
• The novel kernel methodology enhances accuracy by encoding complex correlations and satisfying physical principles.
• This technique holds significant potential for improving clinical applications requiring vascular flow analysis.