Generative Reconstruction of Multimodal Cardiac Waveforms From a Single Vibrational Cardiography Sensor

Area of Science:

• Biomedical Engineering
• Cardiovascular Monitoring
• Signal Processing

Background:

• Multimodal cardiac monitoring systems typically require numerous sensors, limiting their practicality for continuous, everyday use.
• Existing systems face challenges in real-world applications due to complexity and the need for frequent calibration.
• There is a need for simplified, wearable solutions for comprehensive cardiac assessment.

Purpose of the Study:

• To develop and validate a generative modeling framework for estimating multiple cardiac waveforms from a single vibrational cardiography (VCG) sensor.
• To assess the feasibility of using VCG and generative AI to replace complex multimodal sensor setups for cardiac monitoring.
• To improve the practicality of continuous cardiac monitoring in daily life.

Main Methods:

• Recorded VCG signals alongside electrocardiography (ECG), impedance cardiography (ICG), non-invasive blood pressure (NIBP), and photoplethysmography (PPG) in 20 subjects.
• Utilized a conditional Generative Adversarial Network (cGAN) to reconstruct normalized ECG, ICG, NIBP, and PPG signals from VCG inputs.
• Employed a leave-one-subject-out cross-validation strategy to evaluate model generalization and calibration-free performance across diverse physiological states (e.g., breath holding, cold pressor test).

Main Results:

• Reconstructed cardiac waveforms showed strong alignment with target signals, with median Pearson's correlation coefficients ranging from 0.808 (ECG) to 0.929 (PPG).
• The generative model accurately captured both morphological structure and temporal dynamics of the estimated waveforms.
• Accuracy remained consistent across various physiological interventions, demonstrating robust performance.
• Fiducial point analysis confirmed the model's ability to extract key cardiac features.

Conclusions:

• A single-sensor VCG approach combined with generative modeling offers a viable and streamlined alternative to conventional multimodal cardiac monitoring systems.
• This method significantly enhances the potential for practical, continuous cardiac monitoring in everyday settings.
• The developed framework demonstrates the power of AI in simplifying complex physiological signal acquisition.