Development and evaluation of an arterial pulse waveform analysis algorithm

Abstract

Relaxation techniques such as yoga or mantra breathing are known to decrease blood pressure. This particularly applies to guided slow breathing which improves the arterial baroflex sensitivity and thus employs a blood pressure lowering effect. Novel wearable sensor technology enables ubiquitous, continuous and non-invasive measurement of the blood pressures evolution over time by using the pulse transit time as surrogate. To further investigate the potential of pulse waves as indicator for alteration of blood pressure, this thesis deploys a newly devised algorithm using MATLABc to quantify and evaluate pulse waves. In an exhaustive literature research, the state of the art quantification methods were collected and described. From this compilation, several methods were selected for implementation. The employed algorithm can be divided into three major parts. In the first part, the data is preprocessed. A central moving average filter is applied to the signals and the pulse waves are separated. Subsequently, a strict artefact annotation filter rejects non-physiological waves. In part two, the Systolic Pressure, the Dicrotic Notch Pressure and the Dicrotic Wave Pressure of all physiological waves are automatically annotated. Initially, existing algorithms were used and the annotated points were visually examined. The examination of the annotated Dicrotic Notch Pressure yielded an unsatisfactory accuracy for the data sets analysed. Thus, a novel algorithm was designed to find the Dicrotic Notch Pressure. Besides the pressure values, their timings were extracted and analysed. Moreover, derived parameters such as amplitude ratios were calculated to further describe the waves. In the last step, the annotated and derived features were statistically assessed and their timely evolution were depicted. The algorithm was applied to biosignal recordings of 30 subjects with essential hypertension, each lasting for approximately 15 minutes. During the initial ten minutes, the subjects exercised guided breathing followed by five minutes of unguided breathing. Exemplarily, the 3-min averaged duration between the waves onset and the Dicrotic Notch significantly decreased by 5.57 ms (p < 0.01) during the entire recording and with respect to the baseline, implying a leftward shift of the Dicrotic Notch. The Total Pulse Duration significantly decreased by 18 ms (p < 0.05) during the course of the recording, implying a decline in width. This thesis presents an algorithm to automatically separate, filter and annotate pulse waves. The obtained results are promising and encourage further investigation.