Automatic Artifact Component Removal Using a Neural Network in MCG Signal

Abstract

An algorithm combining a neural network and a principal component analysis (PCA) is proposed to remove a pulse-type artifact which often occurs in the 61 channel MCG system installed at Samsung Medical Center in Seoul, Korea. In the proposed work, the acquired signal is first decomposed into components by the PCA, and the components corresponding to the artifact are identified and removed by the neural network. The neural network is an essential component in the automation procedure. Unlike existing artifact rejection algorithms, the proposed algorithm is on a component-by-component basis, and the restored signal is used for further processing once the artifact components are successfully removed. Seven parameters are extracted from each time-domain component and are used as the input to the neural network. They are maximum, minimum, peak-to-peak value, variance, mean, skewness, and kurtosis. In the experiments with volunteers, 97% of the decisions made by the neural network are identical to those by the human experts. Using the proposed technique, the artifact was successfully removed from the MCG signal.