A method of parameter estimation for cardiovascular hemodynamics based on deep learning and its application to...

Area of Science:

• Cardiovascular physiology
• Biomedical engineering
• Computational biology

Background:

• Precise cardiovascular model personalization is crucial for clinical applications.
• Traditional parameter estimation methods can be time-consuming and complex.
• Deep learning offers a promising approach for automating complex physiological modeling.

Purpose of the Study:

• To develop and evaluate a deep learning model for cardiovascular hemodynamics parameter estimation.
• To investigate the use of transfer learning to enhance model personalization.
• To assess the accuracy and efficiency of the proposed deep learning approach.

Main Methods:

• A multi-input deep neural network (DNN) combining convolutional neural networks (CNNs) and fully connected neural networks (FCNNs) was developed.
• The DNN processes pressure waveforms, heart rate (HR), and pulse transit time (PTT) measurements.
• Transfer learning (TL) was employed to improve the personalized characteristics of the trained network.

Main Results:

• The DNN model accurately estimated cardiovascular hemodynamics parameters from both synthetic and in vitro data.
• Transfer learning significantly improved the personalization of the model for individuals with different characteristics.
• A multicycle combination strategy notably enhanced estimation accuracy.

Conclusions:

• The proposed deep learning method provides an immediate, accurate, and sustainable approach for cardiovascular model personalization.
• This technique holds significant potential for advancing cardiovascular research and clinical practice.
• Further attention and development of this deep learning approach are warranted.