Comparison of artificial neural network based ECG classifiers using different features types

Abstract

Artificial neural networks (ANN) have been applied for some years in the field of signal classification with the aim of outperforming the traditional classifiers.This paper addresses the results of a study that comprehended the design and training of ANNs for ECG classification in four classes. Distinct ANNs having as inputs distinct ECG features types were designed and trained with the aim of attaining a reduced and "best" discriminating features set.

Morphograms: exploiting correlation patterns to efficiently identify clinically significant events in intensive care units

“…We shall describe the results of using this method on a publicly available database [5] to detect artifacts in arterial blood pressure (ABP) signals by exploiting correlations with ECG signals. Numerous systems have been developed for automatically analyzing patient monitoring signals, and these have employed various methods ranging from traditional signal processing techniques such as frequency analysis, timefrequency analysis, and wavelet analysis [3], [7], [8], [9], [11] to techniques developed by Artificial Intelligence (AI) researchers such as neural networks (NN) [4], [6], [12], [15] and fuzzy logic [1], [10], [16]. The majority of these systems have focused on ECG signals and most of the systems have used a single-signal model, or when they do look at multi-channels, they are of the same type of signal -ECG signals from multiple leads.…”