Parameters Extraction and Monitoring in Uterine EMG Signals. Detection of Preterm Deliveries Dima ALAMEDINE 1, 2 , Mohamad KHALIL 2 , Catherine MARQUE 1 1 Université de Technologie de Compiègne, BioMécanique et BioIngénierie (BMBI) - UMR CNRS 7338, Neuromécanique et Signaux Electrophysiologiques (NSE), Compiègne, France. 2 Université libanaise-EDST-Centre Azm pour la recherche en biotechnologie et ses applications, Laboratoire des Systèmes électroniques Télécommunications et Réseaux (LaSTRe), Tripoli, Liban. Abstract- The aim of this paper is to classify between Labor contractions and pregnancy contractions. Various types of parameters have been extracted from the electrohysterogram (EHG), mainly from the whole EHG or from different frequency bands. They have been computed from different signal databases obtained with different recording protocols. The results of these studies are sometime controversial. The aim of this paper is to compute 17 parameters selected from the literature on the same signal database, either on the whole EHG or after wavelet packet decomposition, and then to compare their power to discriminate between contractions recorded during pregnancy and labor. We thus obtain a selection of parameters that allow the best discrimination between pregnancy and labor contractions, when computed on the same signals, either on the whole EHG or on selected frequency bands. Index Terms Preterm labor, EHG, parameters extraction, Jeffrey divergence methods. I. INTRODUCTION Preterm birth remains a major problem in obstetrics; it remains the leading cause of neonatal morbidity and mortality. In the most developed country, the World Health Organisation (WHO) reported that the perinatal mortality rate is around 7 per 1,000 births [1]. The socio- economic consequences of these prematurity are important [2]. Indeed some days more in utero can improve the maturation of the fetus and hence its viability at birth [3]. For this we need a reliable method for the early detection and prevention of preterm birth threats. A new method of preventing preterm birth started since the 80s based on the study of uterine electromyographic signal (EHG) [4]. This signal is obtained noninvasively by using surface electrodes attached to the pregnant woman's abdomen. This signal represents the summation of the electrical activity generated by the active uterine muscle cells, corrupted by surrounding electrical and mechanical activities [5]. Several parameters have been extracted from the EHGs, in order to find specific information useful to detect preterm birth. In our study we selected 17 linear parameters which are obtained from the literature. The selected parameters are: mean frequency [6], Peak Frequency [7-8], deciles which contain the median frequency [8-9-10], parameters extracted from wavelet decomposition [11] and parameters extracted from wavelet packet decomposition [12]. On these multiple studies, the parameters being computed from different signal databases, obtained with different recording protocols, it is sometime difficult to compare their results in order to choose the "best" parameter for preterm labor detection. The aim of this paper is to compare these linear parameters, computed on the same EHG signals corresponding to a standardized recording database [13], in order to find the most pertinent parameters which can discriminate between pregnancy and labor contractions. This paper is organized as follows: In section II we explain the experimental protocol. Then we present the extracted linear parameters, and we will describe our methodology for selection of the linear parameters. In section III we present the linear parameters selected that permit the best discrimination between labor and pregnancy contractions. II. MATERIEL AND METHODS A. Experimental protocol The EHG was recorded by placing an array of 16 recording electrodes on the woman's abdomen and two reference electrodes positioned on each of her hip. To reduce the common mode noise, we calculate the vertical bipolar signals as shown in Figure1 (Vbi). In our study we use only the bipolar channel Vb7. We recorded EHG on 39 women: 28 during pregnancy (33-39 weeks of gestation) and 11 during labor (39-42 weeks of gestation) and digitized them with a sample frequency of 100 Hz. After a manual segmentation of the EHG, based on tocodynamometer data, we obtained 106 pregnancy EHG bursts and 106 EHG labor bursts. B. Parameters extraction We first compute the 17 parameters from the whole EHG. To calculate the power spectral density (PSD) we used the Welch Periodogram. Then we extract from the PSD the frequency parameters: mean frequency [6], Peak