Invasive arterial blood pressure delineator for cardiopulmonary resuscitation patients during pauses of chest compressions

Abstract

Invasive arterial blood pressure (IBP) monitoring is important to assess patient’s cardiovascular competence and guide clinical treatment. Besides, international resuscitation guidelines in force suggest its use during Cardiopulmonary Resuscitation (CPR), but current automated algorithms for IBP variables computation were not designed for cardiac arrest patients. A lack of knowledge is detected in the automated processing of IBP signal during CPR.The aim of this study was to design algorithms for heartbeat detection and for IBP physiological variable computation during CPR, and compare to state-of-the-art (SoA) proposals. The dataset used consists of 81 out-of-hospital-cardiac-arrest (OHCA) patients and two additional public datasets with hemodynamically stable patients. A set of 377 IBP segments, total duration of 1127 min, were extracted from the OHCA dataset during the pauses of chest compressions. The method includes artifact removing from the in IBP using Stationary Wavelet Decomposition and heartbeat detection in the first difference signal. A multicomponent evaluation and two adaptive thresholds were applied to compute IBP physiological variables.Pulsatile segments with heartbeats were discriminated from pulseless segments with mean (standard deviation) sensitivity(Se)/specificity and positive (PPV)/negative predictive values of 98.8(6.9)/91.6(20.2)% and 97.4(9.7)/98.7(6.1)%, respectively. The heartbeat detection showed 96.1(8.3)% of Se, 96.1(7.6)% of PPV and 95.7(6.4)% of F1-score , with absolute errors of 0.55(2.91)/0.39(4.87)/0.78(6.08)mmHg in systolic, diastolic and pulse pressure values, respectively. The proposed algorithms outperformed SoA solutions with both OHCA and stable patients.