A Least Mean-Square Filter for the Estimation of the Cardiopulmonary Resuscitation Artifact Based on the Frequency of the Compressions

Abstract

Cardiopulmonary resuscitation (CPR) artifacts caused by chest compressions and ventilations interfere with the rhythm diagnosis of automated external defibrillators (AED). CPR must be interrupted for a reliable diagnosis. However, pauses in chest compressions compromise the defibrillation success rate and reduce perfusion of vital organs. The removal of the CPR artifacts would enable compressions to continue during AED rhythm analysis, thereby increasing the likelihood of resuscitation success. We have estimated the CPR artifact using only the frequency of the compressions as additional information to model it. Our model of the artifact is adaptively estimated using a least mean-square (LMS) filter. It was tested on 89 shockable and 292 nonshockable ECG samples from real out-of-hospital sudden cardiac arrest episodes. We evaluated the results using the shock advice algorithm of a commercial AED. The sensitivity and specificity were above 95% and 85%, respectively, for a wide range of working conditions of the LMS filter. Our results show that the CPR artifact can be accurately modeled using only the frequency of the compressions. These can be easily registered after small changes in the hardware of the CPR compression pads.

Shock Advisory System for Heart Rhythm Analysis During Cardiopulmonary Resuscitation Using a Single ECG Input of Automated External Defibrillators

“…The Se value of 90.1% is equal to the one presented by De Gauna et al, 6 but lower than 95.6% and 96.7% reported by Irusta et al 19 and Eilevstjonn et al, 11 respectively. However, Se of 90.1% is high enough to meet the AHA recommendations for minimal performance goal requirements in AEDs set for detection of VFs (>90%).…”

“…15 The superposition of ECG and CC artefacts often resembles ventricular fibrillation/ tachycardia (VF/VT), 1 resulting in both specificity (Sp) and sensitivity (Se) reduction of AED shock advisory systems (SASs). 3,6,11,19 Therefore, the current practice recommends CPR interruption when it is necessary to assess the rhythm 7 thus providing noise-free ECGs as required for a reliable shock/no-shock decision in AEDs. 21 The adverse effects of the ''hands-off'' intervals during the regular AED analyses are associated with intermittent lack of cerebral and myocardial blood flow that was reported to be deleterious to the tissues.…”

“…Part of them are managed by synthesized reference signals, such as the CC frequency value. 3,6,18,19 Others are managed by reference signals acquired via outer sensors for thoracic impedance, 1,11,17 compression acceleration, 11,17 compression displacement, 1 ''ECG common'' signal, 11,17 arterial blood pressure signal, 27 and CC frequency signal. 2,5 Most of these reference signals are not fitted for use in OHCA interventions.…”

Part 5: Adult Basic Life Support

“…In 6 LOE 5 studies [191][192][193][194][195][196] using human-derived ECG recordings with actual or simulated CPR artifacts and 1 LOE 5 study in a swine model of VF, 197 the use of computerized algorithms that removed compression artifacts from the ECG during CPR reduced the accuracy of rhythm analysis relative to rhythm analysis during pauses. Sensitivity was between 90% and 98%, which would cause inappropriate prolongations in chest compression for shockable rhythms in up to 1 out of 10 patients.…”

Direct evaluation of the effect of filtering the chest compression artifacts on the uninterrupted cardiopulmonary resuscitation time

“…Instead, state-of-the-art studies of filtering evaluate the performance of the filters using short-duration (10-20 seconds) OHCA records corrupted by CPR and perform a single rhythm analysis before and after the CPR artifact is filtered. The researchers then evaluate how the sensitivity and specificity of the shock advice algorithm (SAA) changes after filtering [6][7][8][9]. In this context, the performance of current filter-SAA combinations is still not good enough because the specificity after filtering is too low at below 90% in all studies [10].…”