Cardiac parameter estimation from electrocardiograms based on inversion of physical simulator behavior

Area of Science:

• Biomedical Engineering
• Computational Physiology
• Artificial Intelligence in Medicine

Background:

• Electrocardiograms (ECGs) offer rich data on heart states, but identifying abnormality causes is challenging for specialists.
• Current methods rely on template matching, which has limitations in diagnosing complex cardiac conditions.

Purpose of the Study:

• To develop a deep neural network model for estimating microscopic cardiac parameters from 12-lead ECGs.
• To provide insights into internal heart states and electrophysiological factors influencing ECG signals.

Main Methods:

• A deep neural network with an encoder and multiple decoders was designed to convert ECGs into cardiac parameters.
• A white-box heart simulation model was used to generate essential training data pairs of cardiac parameters and ECGs.
• The model's performance was evaluated using mean absolute error for continuous parameters and accuracy for discrete parameters.

Main Results:

• The proposed deep learning model accurately estimated both continuous and discrete cardiac parameters.
• Detailed analysis of estimation errors through visualization confirmed the model's precision.
• The method demonstrated potential for automatic cardiac parameter estimation from ECGs.

Conclusions:

• The developed model shows promise for clinical applications in monitoring heart conditions and understanding ECG-ECG waveform-disease relationships.
• Further validation with real-world ECG data is necessary before clinical implementation.