Asymptotic-Preserving Neural Networks for Viscoelastic Parameter Identification in Multiscale Blood Flow Modeling
A new neural network can infer arterial pressure from simple ultrasound scans, potentially replacing invasive catheters.
Researchers Giulia Bertaglia and Raffaella Fiamma Cabini have developed a novel AI framework called Asymptotic-Preserving Neural Networks (APNNs) that addresses a critical challenge in cardiovascular modeling: determining the viscoelastic properties of arterial walls without invasive procedures. Their approach embeds the governing physical principles of multiscale blood flow models directly into the neural network's learning process, creating a physics-informed machine learning system that can reliably infer parameters controlling how arteries deform under pulsatile pressure.
Unlike traditional methods that require complex, invasive measurements, this APNN framework can estimate pressure waveforms using only readily available patient-specific data from Doppler ultrasound scans—specifically cross-sectional area and velocity measurements. The system simultaneously reconstructs the time-dependent evolution of blood vessel state variables while identifying the viscoelastic parameters, essentially solving two problems at once. This dual capability makes it particularly valuable for vascular segments where direct pressure measurements are impractical or impossible to obtain.
The methodology has been validated through extensive numerical simulations in both synthetic scenarios and actual patient-specific cases, demonstrating effectiveness across different vascular conditions. By bridging the gap between mathematical models and clinical applications, this research represents a significant advancement in non-invasive cardiovascular diagnostics, potentially reducing the need for catheter-based pressure measurements while providing clinicians with more accessible tools for assessing arterial health and disease progression.
- APNNs embed physical blood flow principles directly into neural network training for accurate parameter identification
- Estimates pressure waveforms from standard Doppler ultrasound data (area + velocity) without invasive catheters
- Validated in both synthetic simulations and patient-specific scenarios showing clinical applicability
Why It Matters
Could transform cardiovascular diagnostics by providing critical pressure data non-invasively, reducing patient risk and expanding monitoring capabilities.