Detection of ventricular Arrhythmias using roots location in AR-modelling

This paper addresses the problem of automatic discrimination of rhythms in ECG signals. In performing the discrimination, fourth-order AR parameters of successive segments are estimated and the related roots are computed and used as inputs to the learning vector quantization (LVQ) classification algorithm. In discriminating normal (NSR) rhythm from arrhythmias, 98% of normal data and 88% of data with arrhythmia are classified correctly. Also in discriminating VF from VT, 72% of data with VF and 86% of data with VT are determined properly.