Central aortic pressure waveform estimation from photoplethysmogram using variational mode decomposition and...

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Artificial Intelligence in Medicine

Background:

• Central aortic pressure waveform (CAPW) is crucial for cardiovascular assessment.
• Current methods often require invasive procedures or calibration.
• Non-invasive estimation of CAPW remains a significant clinical challenge.

Purpose of the Study:

• To develop and validate a novel, calibration-free method for estimating CAPW directly from photoplethysmogram (PPG) signals.
• To integrate signal processing techniques with a sophisticated hybrid neural network for accurate CAPW reconstruction.
• To assess the clinical applicability of the proposed method through extensive validation.

Main Methods:

• Proposed a calibration-free approach utilizing variational mode decomposition (VMD).
• Developed a hybrid neural network combining temporal convolutional network (TCN), gated recurrent unit (GRU), and self-attention mechanisms.
• Validated the method on a large dataset of 4374 virtual subjects using subject-level splitting.

Main Results:

• Achieved excellent accuracy in estimating CAPW indices compared to gold-standard values.
• Reported low mean absolute errors (1.99-3.17 mmHg) and high coefficients of determination (0.89-0.98).
• Demonstrated minimal mean differences (-0.39-0.12 mmHg), indicating high agreement with reference measurements.

Conclusions:

• The developed method provides a promising non-invasive and calibration-free approach for CAPW estimation.
• This technique has the potential to significantly advance cardiovascular assessment tools.
• The study highlights the efficacy of integrating VMD with hybrid neural networks for complex physiological signal analysis.