Classification of Cardiac Rhythms During Load-Distributing Band Cardiopulmonary Resuscitation

Rhythm analysis is crucial during cardiopulmonary resuscitation (CPR) in order to make decisions about therapy. Nowadays, chest compressions (basic treatment for cardiac arrest) must be stopped for reliable rhythm analysis, because of the artefact they induce in the ECG signal, jeopardizing the favorable outcome of the patient. In this work, a method to discriminate between asystole (AS), organized (OR) and shockable (Sh) rhythms is proposed during ongoing mechanical chest compressions using Load Distributed Band (LDB, Autopulse) device during out-of-hospital cardiac arrest. The artefact was reduced using an adaptive filter, 83 features were extracted based on stationary wavelet transform and a random forest classifier was applied. The unweighted mean of sensitivities in the 10-fold cross-validation was 84.3% and mean F1-score was 83.4%. Besides, the algorithm met American Heart Association requirements, with a sensitivity of 91.0% and a specificity of 97.2% for Sh class.