Classification of fetal movement accelerometry through time-frequency features

Abstract

This paper presents a time-frequency approach for fetal movement monitoring which is based on classification of accelerometry signals collected from pregnant women's abdomen. Features extracted from time-frequency distribution of these signals were supplied into statistical analysis to generate feature-measure mixtures. Four various classes subjectively are recognized in accelerometry data by means of objective tools such as ultrasound sonography. These include strong and weak fetal movement, artefact, and background. Receiver operating characteristic analysis utilized to compute the performance of feature-measures for the comparison between various classes. Next, a feature selection applied to reduce the feature space dimension by means of principal component analysis. The selected feature-measures then employed in support vector machine classifiers to classify artefact and fetal movement in different subsets of available classes. The results indicate the fetal movement events are identified with an accuracy of 92.19%.