Estimation of Quasi-Continuous Blood Pressure based on Harmonic Phase-Shifts in Signals using Non-Invasive Photoplethysmographic Measurements

According to the guidelines of the European Society of Hypertension International Protocol revision 2010, the requirements for long time blood pressure measurement (BPM) are: simple handling, robustness against movements, accuracy of better than 5 mmHg, and, above all, that the patient9s motion should not be restricted during measurement. These requirements are in particular important for a reliable interpretation of the blood pressure (BP) of hypertensive patients, because such a diagnosis will usually be confirmed by long-term measurement. Moreover, to be able to correlate the patient9s BP with his normal daily activity, non-obstructive non-invasive methods are desired to reduce the patient9s load. The main concern of this paper is to present a novel method for estimating non-invasive continuous blood pressure (CBP) from a single photoplethysmography (PPG) signal. In contrast to the pulse transit time (PTT) method, our approach is based on the assumption that the phase-velocities of the fundamental and higher harmonics depend on the (non-linear) elastic properties of the arteries. Consequently, phase velocity varies as a function of a vessel9s instantaneous dilation and can be effectively utilised for CBP estimation. In addition to its numerous advantages for a simplified measurement setup, we could show that the method achieves a high degree of correlation for a reliable BP estimation from PPG data. Comparison with state-of-the-art PTT methods was carried out using a dataset from the PhysioBank Database comprising a reference invasive blood pressure (IBP) signal measured at the radial artery, a PPG signal measured at the fingertip and a standard ECG signal. The correlation values obtained from the long-time estimation of the systolic blood pressure (SBP) were as high as r = 0.8945, while the value for the diastolic blood pressure (DBP) was found to be r = 0.9082 and the correlation of the mean blood pressure (MBP) was r = 0.9322. These results were achieved by analysing the dataset in a beat-to-beat manner and regarding several post-processing procedures like coherent averaging (CA) and zero padding with quasi-continuous frequency domain estimation and artificially refined frequency resolution.