On-line Learning Algorithms for extracting respiratory activity from Single Lead ECGs based on Principal Component Analysis

Abstract

In this paper we present several statistic gradient algorithms from literature to solve the Principal Component Analysis (PCA) problem. We used a linear artificial neural network forming the basis of the implemented algorithms which is a neat way for on-line computation of the PCA expansion. As convergence is a key-aspect of these algorithms and is crucial for the usefulness in particular applications, we compared the different learning rules with respect to their suitability in ECG signal processing. Recent studies have shown, that a surrogate respiratory signal can be derived from single-lead ECGs by applying PCA. Since the traditionally applied closed-form computations of PCA are numerically demanding, it seems promising to resort to an adaptive approach when dealing with changing environments like the ECG.

Online learning algorithms for principal component analysis applied on single-lead ECGs

“…In [ 18 ], we have evaluated different learning algorithms in an attempt to extract respiratory activity out of single-lead ECGs acquired from the Physionet Database. As the results were quite promising, we decided to develop a new measurement system consisting of a wireless multichannel ECG sensor-module in combination with a piezo respiration belt sensor-module.…”