Continuous Blood Pressure Measurement Guided by Physiological Mechanisms

Photoplethysmography (PPG) technology is widely used in wearable devices for cuffless blood pressure (BP) estimation. However, the accuracy of BP estimation still needs improvement due to the complex factors influencing BP formation. This study introduces a PPG-based BP estimation method guided by physiological mechanisms. Our approach effectively captures key features related to systolic (SBP) and diastolic blood pressure (DBP) by designing BP-oriented attention mechanisms. We enhance the detection of subtle changes using a Siamese pyramid network architecture and employ a Kolmogorov-Arnold Network (KAN) regression head to better map the complex relationship between PPG and BP. Tested with 361 subjects from VitalDB dataset, our approach achieved an SBP prediction accuracy of -0.31 ± 5.90 mmHg and a DBP prediction accuracy of 0.05±3.29 mmHg, complying with the AAMI blood pressure standard. Under the BHS standard, SBP reaches grade B and DBP reaches grade A. The effectiveness of each module is validated through ablation experiments. This method enhances the accuracy of predictions for SBP and DBP with high BP variability, offering new possibilities for non-invasive BP monitoring.