Wavelet-based event detection in implantable cardiac rhythm management devices.

Abstract

This paper presents a novel event detector for implantable devices. The algorithm is based on a signal model which describes an event as a linear combination of basis functions. The linear combination involves two fundamental electrogram waveforms represented at different time scales. An efficient, low-complexity detector is developed using the dyadic wavelet transform with integer filter coefficients, and a generalized likelihood ratio test. The results show that reliable detection is obtained at an intermediate signal-to-noise ratio (SNR = 25 dB) for various common noise sources. In terms of probabilities of missed events and false alarms, an over-all performance of 0.7% and 0.1%, respectively, was achieved on electrograms corrupted by the different noise types at an intermediate SNR.