Classifier architectures for single chamber arrhythmia recognition

摘要

In this paper we study heart arrhythmia classification for single chamber implantable cardio-verter defibrillators. Our research shows that performance of conventional classification methods using only simple heart rate timing based features can be improved with the inclusion of morphology analysis on samples of the right ventricular apex lead. While morphology classification is typically patient dependent, and computationally expensive, we show that the performance of a patient independent classifier which uses a multi-layer perceptron for morphology recognition and heart beat timing decision tree is superior to that of a timing only classifier, while remaining economical (silicon area and power dissipation) from an implementation perspective. We also show that performance can be significantly improved in the patient dependent case.