ECG Rhythm Analysis During Manual Chest Compressions Using an Artefact Removal Filter and Random Forest Classifiers

Abstract

Interruptions in cardiopulmonary resuscitation (CPR) decrease the chances of survival. However, CPR must be interrupted for a reliable rhythm analysis because chest compressions (CCs) induce artifacts in the ECG. This paper introduces a double-stage shock advice algorithm (SAA) for a reliable rhythm analysis during manual CCs. The method used two configurations of the recursive least-squares (RLS) filter to remove CC artifacts from the ECG. For each filtered ECG segment over 200 shock/no-shock decision features were computed and fed into a random forest (RF) classifier to select the most discriminative 25 features. The proposed SAA is an ensemble of two RF classifiers which were trained using the 25 features derived from different filter configurations. Then, the average value of class posterior probabilities was used to make a final shock/no-shock decision. The dataset was comprised of 506 shockable and 1697 non-shockable rhythms which were labelled by expert rhythm resuscitation reviewers in artifact-free intervals. Shock/no-shock diagnoses obtained through the proposed double-stage SAA were compared with the rhythm annotations to obtain the Sensitivity (Se), Specificity (Sp) and balanced accuracy (BAC) of the method. The results were 93.5%, 96.5% and 95.0%, respectively.

Automated Condition-Based Suppression of the CPR Artifact in ECG Data to Make a Reliable Shock Decision for AEDs during CPR

“…The performance of the algorithm was further evaluated on an independent validation data set. Figure 6 and Table 3 indicate that the defined threshold value for the spectral power in the frequency band (10)(11)(12)(13)(14)(15) Hz is also valid for the unseen ECG data. Our analysis showed that nearly 80% of the non-shockable samples have total spectral power higher than the defined threshold value while only 12% of the shockables breach this threshold.…”

“…Various filtering methods using numerous signal processing techniques have been developed during the last two decades to suppress CPR artifacts [5,6,8]. The majority of studies are based on Kalman filters [9,10], different adaptive filtering methods such as least mean square (LMS) [11,12], the enhanced adaptive method [13,14], and recursive least squares (RLS) [15,16]. Almost all of the above-noted methods require one or more reference signals (such as chest pressure, chest displacement, chest acceleration, compression depth, or thoracic impedance).…”

“…Estimated power (watt/Hz) in the validation samples for frequencies within(10)(11)(12)(13)(14)(15) Hz.…”