An unsupervised classification method of uterine electromyography signals: Classification for detection of preterm deliveries

Abstract

This article proposes an unsupervised classification method that can be applied to the electromyography signal of uterine contractions for the detection of preterm birth. The frequency content of the electromyography changes from one woman to another, and during pregnancy, so wavelet decomposition is first used to extract the parameters of each contraction, and an unsupervised statistical classification method based on Fisher's test is used to classify the events. A principal component analysis projection is then used as evidence of the groups resulting from this classification. Another method of classification based on a competitive neural network is also applied on the same signals. Both methods are compared. Results show that uterine contractions may be classified into independent groups according to their frequency content and according to term (either at recording or at delivery).