Related Experiment Video
Updated: Jul 26, 2025
08:22
Author Spotlight: Advancing the Study of Brain-Heart Interplay with a Comprehensive EEGLAB Plugin for Multimodal Signal Analysis
Published on: April 26, 2024
1.9K
Time-series NARX feedback neural network for forecasting impedance cardiography ICG missing points: a predictive
Sara Benouar1,2, Malika Kedir-Talha2, Fernando Seoane1,3,4,5
1Department of Clinical Science, Intervention and Technology, Karolinska Institutet, Stockholm, Sweden.
Frontiers in Physiology
|June 22, 2023
Summary
This study introduces a predictive model using a NARX neural network to forecast missing characteristic points in impedance cardiography (ICG) signals. This significantly improves the detection accuracy of crucial hemodynamic parameters from ICG data.
Area of Science:
- Biomedical Engineering
- Cardiovascular Physiology
- Artificial Intelligence in Medicine
Background:
- Accurate hemodynamic parameter assessment via impedance cardiography (ICG) relies on detecting characteristic points (X, Y, O, Z) in the dZ/dt complex.
- Variability in ICG complex subtypes challenges beat-to-beat accuracy for parameters like stroke volume and cardiac output, particularly the left ventricular pre-ejection time.
- Existing methods struggle with detecting these critical points, leading to significant data loss and reduced reliability.
Purpose of the Study:
- To develop a predictive model capable of accurately forecasting missing characteristic points in diverse ICG complex subtypes.
- To enhance the detection rates of ICG characteristic points, thereby improving the extraction of vital hemodynamic parameters.
- To reduce the percentage of missing data points in ICG analysis, supporting more robust beat-to-beat calculations.
Main Methods:
- Implementation of a time-series non-linear autoregressive with exogenous inputs (NARX) feedback neural network.
- Training the NARX model on two distinct datasets in an open-loop configuration for accurate feedback.
- Conducting multi-step prediction tests and simulations after successful model training with closed-loop functionality.
Main Results:
- The NARX model achieved a success rate of 75%-88% in predicting missing characteristic points across evaluated ICG datasets.
- This represents a substantial improvement compared to previous detection rates of 21%-30% without the predictive model.
- The model demonstrated effectiveness in predicting X, Y, O, and Z points across different ICG complex subtypes.
Conclusions:
- The developed NARX-based predictive model offers a promising approach for improving ICG data analysis.
- Significant accuracy increases in detecting characteristic ICG points are achievable, enhancing hemodynamic parameter extraction.
- This method supports more reliable beat-to-beat hemodynamic monitoring by mitigating issues related to ICG complex variability.
