Detection of Shockable Rhythm during Chest Compression based on Machine Learning

Recently, more and more evidence has shown that reducing the interruption of chest compression to increase the duration of blood flow perfusion is the key to improve the successful rate of CPR. Many researchers have been devoted to designing algorithms that can identify shockable ECG rhythms during chest compression. Most of these studies focused on designing filters that can eliminate CPR artifacts from the corrupted ECG signals. Although some adaptive filtering algorithms have suppressed the artifacts of chest compression to a great extent, the residual artifacts of CPR still can not be ignored. In this paper, we constructed a robust shockable rhythm detection algorithm based on machine learning, which can distinguish accurately different types of ECG signals even under the condition of severe CPR artifacts interference. A total of 21 metrics were extracted from the ECG signals by a large number of retrospective studies of the existing shockable detection algorithms. After feature selection, 13 metrics were selected to participate in BP neural network construction. The performance of this network was evaluated on constructed corrupted ECG signals, sensitivity and specificity could be maintained above 99% and 95% respectively even at a particularly high corruption level.