Bidirectional Translation Between ECG and PCG.

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Simultaneous electrocardiography (ECG) and phonocardiogram (PCG) provide complementary insights into cardiac electrical and mechanical function.
• Understanding the interplay and reconstructability between ECG and PCG signals is crucial for advanced cardiac monitoring.

Purpose of the Study:

• To investigate the shared and unique information between ECG and PCG signals.
• To evaluate the feasibility of reconstructing one signal modality from the other using various models.
• To assess the performance of linear and nonlinear models, particularly neural networks, for signal reconstruction.

Main Methods:

• Analysis of the EPHNOGRAM dataset containing simultaneous ECG-PCG recordings during rest and exercise.
• Application of linear and nonlinear modeling techniques, including a non-causal neural network, for signal reconstruction.
• Quantitative evaluation of reconstruction performance using metrics like signal-to-noise ratio (SNR) and cross-correlation.

Main Results:

• Nonlinear models, especially the non-causal neural network, demonstrated superior performance in reconstructing cardiac signals.
• Reconstruction of ECG from PCG was found to be more feasible than the inverse.
• The non-causal neural network achieved an SNR of 6.5±5.2 dB and a cross-correlation of 0.78 ± 0.19 for PCG-based ECG reconstruction in a within-subject analysis.

Conclusions:

• The study quantifies the electromechanical relationship between cardiac electrical and mechanical signals.
• Non-causal neural networks show significant potential for reconstructing ECG from PCG data.
• Findings support the development of novel multimodal cardiac monitoring systems leveraging both ECG and PCG signals.