Central aortic pressure waveform estimation from photoplethysmogram using variational mode decomposition and TCN-GRU-SAM

Central aortic pressure waveform (CAPW) provides critical clinical insights for cardiovascular assessment. This study proposed a calibration-free method to estimate CAPW directly from photoplethysmogram by integrating variational mode decomposition with a hybrid neural network combining temporal convolutional network, gated recurrent unit, and self-attention mechanism. Validation on 4374 virtual subjects via subject-level splitting demonstrated excellent accuracy of the estimated indices against gold-standard values, as evidenced by mean absolute errors of 1.99-3.17 mmHg, coefficients of determination of 0.89-0.98, and mean differences of -0.39-0.12 mmHg. This method provides a promising step toward clinically applicable, calibration-free CAPW estimation.