A machine learning algorithm to improve patient-centric pediatric cardiopulmonary resuscitation

Abstract

Several studies suggest that the outcome of pediatric cardiopul-monary resuscitation depends strongly on timely recognition of the underlying cause for cardiac arrest, the prominent ones being primary ventricular fibril-lation and secondary asphyxia-associated ventricular fibrillation. If the cause could be determined within the first minute of cardiopulmonary resuscitation, the administration could be appropriately modified in order to achieve optimal outcome. However, distinguishing the two causes has been a difficult challenge. To derive a robust algorithm with acceptable accuracy that distinguishes primary ventricular fibrillation and secondary asphyxia-associated ventricular fibrillation within the first minute of starting cardiopulmonary resus-citation. We address this problem with MACWAVE, a new computa-tional technique integrating advanced signal processing and machine learning. MACWAVE is an algorithm that uses wavelet transforms with electrocardiog-raphy data to identify the most differentiating characteristics of the signal and uses them as features to develop a support vector machine classification model. The developed algorithm shows an average classification accuracy of 85%, this being the first result ever achieved for this critical pediatric problem. Being the first research effort to ever analyze this critical pediatric problem, the MACWAVE method can improve patient-centric cardiopulmonary resuscitation treatment and significantly increase positive outcomes for pediatric cardiac arrest.