Core Concepts
Proposing a novel physics-informed DeepONet framework for continuous ABP waveform prediction.
Abstract
The article introduces the BP-DeepONet method for cuffless blood pressure estimation using a physics-informed DeepONet approach. It addresses the importance of continuous ABP waveform prediction for cardiovascular health and proposes a novel framework that outperforms traditional methods. The study focuses on predicting ABP waveforms at different locations in the artery, incorporating the Navier-Stokes equation and Windkessel boundary condition. Meta-learning is introduced to estimate model hyper-parameters during training, showcasing the effectiveness of the approach through numerical experiments.
Introduction:
Cardiovascular diseases are a leading cause of death globally.
Importance of monitoring ABP waveforms continuously.
Data Extraction:
"In 2019, about 32% of all global deaths were due to CVDs."
"The proposed method outperforms traditional approaches in predicting ABP waveforms."
Physics-informed Machine Learning for PDEs:
Two main directions: learning an instance of PDEs and learning the solution operator of PDEs.
Introduction to physics-informed neural networks (PINNs) and neural operators for PDEs.
BP-DeepONet:
Structure of the BP-DeepONet with branch net and trunk net.
Efficient implementation for training the BP-DeepONet.
Meta BP-DeepONet:
Introduction of sample-dependent hyper-parameters estimation.
Training the hyper-network on top of the branch net.
Numerical Experiments:
Validation of the proposed physics-informed training method using a simulated PDE instance.
Comparison of the PINN solution with a referenced numerical solution.
Stats
"In 2019, about 32% of all global deaths were due to CVDs."
"The proposed method outperforms traditional approaches in predicting ABP waveforms."
Quotes
"Cardiovascular diseases are a leading cause of death globally."
"The proposed method outperforms traditional approaches in predicting ABP waveforms."