Introducing Grammatical Evolution for FHR analysis and classification Ioannis Tsoulos, Georgoulas Georgoulas, Dimitris Gavrilis, Chrysostomos Stylios Member, IEEE, Joao Bernardes, Peter Groumpos Senior Member, IEEE Abstract⎯Electronic fetal monitoring is an essential tool for fetal surveillance during labour. It is mainly based on the monitoring and evaluation of the Fetal Heart Rate, (FHR) which is a biosignal that has to be interpreted on line. Evaluation and interpretation of FHR gives an indication of the fetus health status. A lot of research efforts have been done towards the development of automatic and reliable methods for processing and evaluating FHR. This research work, introduces an integrated methodology for processing and classifying FHR based on the novel approach of grammatical evolution for feature construction and selection. The proposed methodology is presented, which then is applied to data set. Experimental results are promising paving the way for further research in that direction. Index Terms⎯Fetal heart rate, Genetic algorithms, Grammatical evolution, Hypoxia, Neural Networks, SMOTE I. INTRODUCTION The main mean for antepartum and intrapartum fetal surveillance is the Electronic Fetal Monitoring (EFM). The EFM is based on the continuous recording and monitoring of the instantaneous Fetal Heart Rate (FHR) (beats/min) and Uterine Activity (UA), which is also called cardiotocogram (CTG). The typical printout of a CTG consists of the FHR at the upper part and the UA at the lower part for the same time axis (Figure 1). Obstetricians use the CTG during the crucial period of labor to monitor the fetal condition so as to avoid neonatal compromise, namely metabolic acidosis [1]. The medical device that is used for acquiring, processing, displaying and printing out the FHR and UA signals is the cardiotocograph. Despite the fact that EFM was introduced more than four decades ago, there is still controversy regarding its effectiveness, especially among obstetricians. Another I. Tsoulos is with the Computer Science Department, University of Ioannina, 45100 Ioannina, Greece (e-mail: sheridan@cs.uoi.gr ). G. Georgoulas is with the Laboratory for Automation and Robotics, Department of Electrical and Computer Engineering, University of Patras, Patras, 26500, Greece (telephone: 30-610-997293, e-mail: georgoul@ee.upatras.gr). D. Gavrilis is with the Department of Electrical and Computer Engineering, University of Patras, Patras, 26500, Greece ( e-mail: gavrilis @ee.upatras.gr). C. D. Stylios, was with Laboratory for Automation and Robotics. He is now with the Depart. of Communications, Informatics and Management, TEI of Epirus, 47100 Artas, Greece (email: stylios@teleinfom.teiep.gr ) J. Bernades is with the Department of Gynecology and Obstetrics, Porto Faculty of Medicine, Porto, Portugal Email: jbernardes@mail.telepac.pt P. P. Groumpos is with the Lab. for Automation & Robotics, University of Patras, Patras 26500 Greece (telephone: 30-610-997295, e- mail: groumpos@ee.upatras.gr). important issue is that statistical studies revealed an increase in operative vaginal deliveries when pregnant women are monitored with EFM during the intrapartum period [2]. Furthermore, studies on the FHR analysis and interpretation by obstetricians have shown significant inter- observer and intra-observer variation in tracing interpretation [3]. Even though specific guidelines have been published for FHR interpretation [4],[5], the different levels of experience of the various specialists, along with the subjectivity of the approach, have great influence on their final judgment. All these facts have created a mistrustful environment for FHR monitoring and interpreting methods. In addition, the difficulty in distinguishing benign variant patterns from patterns associated with significant fetal acidemia may have arisen because FHR monitoring was introduced into clinical practice before the physiological mechanism that defines the FHR patterns was well understood. 0 5 10 15 20 25 30 30 60 90 120 150 180 FHR beats/min 80 0 5 10 15 20 25 30 35 0 20 40 60 TOCO TOCO % Fig. 1. A simple cardiotocogram (printout) consisting of the FHR at the upper part and UA at the lower part. On the other hand, there is an ongoing interest for more automated methods for FHR processing and analysis that drives the development of computer based systems able to analyse, classify and interpret the CTG [7]-[30]. These approaches are based on classical signal processing methods, Neural Networks, Fuzzy logic and hybrid methods. Some of these efforts tried to develop a system not to just record the CTG, but actually monitor the condition of the fetus in a reliable, effective and reproducible manner. These research efforts have shown that it is still worth to further investigate methods to analyze the FHR not just by imitating the way a clinician does, but by employing techniques based on the signal processing and pattern recognition fields. In this research work we present a new integrated methodology for interpreting, feature extraction and classification of the FHR. The proposed methodology uses the conventional well known method to identify the FHR