Discriminating Pregnancy and Labour in Electrohysterogram by Sample Entropy and Support Vector Machine

Preterm birth (PTB) is a major cause of perinatal mortality and long-term morbidity. The Electrohysterogram (EHG) is non-invasive and a real-time technology used to detect, diagnose or predict the uterine contraction related to PTB. In hospital practice, EHG recordings are obtained and then visually inspected by clinicians for uterine contraction information, which is a very tedious, time-consuming and high-cost task. Therefore, there is a great demand for automatic detection of PTB. This paper presents a novel method for automatic uterine contraction identification based on sample entropy (SampEn) and genetic algorithm-support vector machine (GA-SVM). By using the entropy theory, SampEn values of 16 channels of each sample were calculated to construct feature vectors. Then the identification model was constructed based on SVM with radial basis function (RBF) kernel. These feature vectors were fed into the SVM model for identification. In order to improve the training speed and the accuracy of the identification model, this paper designs a method for obtaining optimal values of penalty parameter and RBF kernel function parameter based on genetic algorithm (GA). In assessing the performance of the identification model, identification accuracy, sensitivity and specificity were considered. Experimental results reveal that compared with the backpropagation (BP) algorithm and SVM, the classification performance of the GA-SVM is better in terms of classification accuracy and specificity which achieves a satisfying recognition results with accuracy of 95.20% and specificity of 96.14%. Finally, experimental results indicate that the method proposed in this work could be effective in identifying the uterine contraction by EHG.