Analysis and Interpretation of Uterine Contraction Signals Using Artificial Intelligence

Abstract

Electrohysterography (EHG) is the technique used to monitor the activity of the uterine signals. The EHG signals are acquired from the abdominal surface of pregnant women, and the readings used to study the electrical activity produced by the uterus. The electrical signal obtained from the abdominal surface helps to differentiate the true labor and false labor pain. EHG signals are recorded from the three channels. The objective of the proposed work is to differentiate the true labor and false labor pain from the EHG signal of the pregnant woman. The proposed work employs EHG signal available in the physionet database. Three channels are used to extract the EHG signal. The dataset consists of 300 records, in which each record consists of three signals recorded using three channels from the abdomen of a pregnant woman. The dataset contains 160 true labor signals and 140 false labor signals. Then obtained signal is filtered by using the five pole order Butter worth band pass filter. The cut-off frequency applied for the butter worth band pass filter is 0.3–1 Hz. The proposed work uses the features such as mean, median, maximum frequency, median frequency, kurtosis, skewness, energy and entropy extracted from the signals for identifying true labor/false labor pain signals. This work employs different classifiers individually for classifying the signals into true labor pain/false labor pain 354signal based on the values of the features extracted. This chapter compares the performance of different machine learning algorithms such as support vector machine (SVM), extreme learning machine (ELM), k-nearest neighbor (KNN), artificial neural network (ANN), radial basis function neural network (RBNN) and random forest (RF) classifiers in identifying true and false labor pain signals. The performance of each classifiers, are evaluated individually. The performance of SVM classifier is evaluated with different kernel functions like linear, polynomial, radial basis function (RBF) and multilayer perceptron (MLP). SVM yields an accuracy of 58%, 55%, 57%, and 55% for different kernel functions linear, polynomial, RBF and MLP respectively. The Performance of KNN classifier is evaluated with kernel function 1 norm. KNN classifier yields an accuracy of 77% for kernel function 1 norm. The Accuracy for different machine learning algorithms such as ANN, RBNN, RF, ELM are 96%, 97%, 82% and 98%, respectively. The highest accuracy of 98% is obtained for ELM classifier. Hence ELM classifier outperforms other classifiers. This chapter proposes the possibility of employing artificial intelligence technique with EHG signals for differentiating true labor and false labor pain signals. Early diagnosis of premature delivery helps to delay the delivery by proper in-time treatment and thus it helps to prevent the premature baby and its associated health issues and risk of death.