A novel two-layered bayesian classifier for atrial tachyarrhythmia

Abstract

The success of implantable cardioverter defibrillator (ICD) led to the concept of a device that would terminate atrial fibrillation (AF). Implantable device for atrial defibrillation are undergoing rapid evolution. Currently used devices combine pacing and cardioversion therapies both to prevent and to treat AF. The success of device therapy for AF depends on rapid and accurate detection of AF, which remains to be a difficult task. Furthermore, low power consumption is equally important for implementing the algorithm to implantable device for AF. Recently, a multi-feature Bayesian classifier was developed and patented. Although it has been successful in accuracy improvement, the design was not optimized to fully utilize the data set information. In this paper, an indepth multi-variate statistical data analysis was performed and a two-layered architecture was proposed. The classification accuracies were further enhanced, from 96.57% to 99.14% at sinus rhythm, from 97.95% to 98.50% at atrial fibrillation and from 95.67% to 96.13% at atrial flutter. The significant increment in sinus accuracy would save precious ICD power. It is concluded that the proposed two-layered classifier can perform better in accuracy by employing less features and the experiment result can provide a solid foundation for designing low-power devices for AF.