Nearest-manifold classification approach for cardiac arrest rhythm interpretation during resuscitation

In order to monitor the cardiac arrest patients response to therapy, there is a need for methods that can reliably interpret the different types of cardiac rhythms that can occur during a resuscitation episode. These rhythms can be categorized to five groups; ventricular tachycardia, ventricular fibrillation, pulseless electrical activity, asystole, and pulse generating rhythm. The objective of this study was to develop machine learning algorithms to automatically recognize these rhythms. We proposed a detection algorithm based on the nearest-manifold classification approach using a group of 8 time-domain features as statistical measures on the signal itself, as well as the first and second differences. The overall accuracy of the cardiac arrest rhythm interpretation is 79% which is 9% better than our prior work. The sensitivity/specificity of shockable/non-shockable rhythms is 92/95%.