Uterus Elektromyogram Sinyalleri Kullanarak Kasılmaların Tespiti Ve Erken Doğum Kestirimi
xmlui.mirage2.itemSummaryView.MetaDataShow full item record
ABSTRACT DETECTION OF CONTRACTIONS AND ESTIMATION OF PRETERM BIRTH BY USING UTERUS ELECTROMYOGRAM SIGNALS AYŞE TAŞDÖĞEN Master, Department of Electrical and Electronics Engineering Supervisor: Assoc. Prof. DR. ATİLA YILMAZ January 2017,100 pages One in ten baby, worldwide borns before 37 weeks of pregnancy. Preterm birth brings lots of problems to the family afterwards. The mother who gives preterm birth and her child must stay inthe hospital for a beginning and the child must be treated carefully.These babies might have many preterm associated discomfortssuch as respiratory insufficiency, lung disorders, weak immune systems, mental problems and learning disordersin their future. For this reason, thecorrect estimation of preterm has a great importance in order to increase the possibilities of eliminating some of those problems. In this study, uterus EMG signals are used to estimate preterm births. In the first part of the study, using some features selectedafter the literature review, specifically contractions which are the set of important events occured on the EMG signals were classified from other significant events. For the classification stage, minimum distance classifiersas a linear classifier example and artificial neural networks as a non-linear classifier example are used. The performance of minimum distance classifier has been improved by adjusting class weights based on known class samples. Artificial neural networks classify the events like contraction, possible contraction and foetal movements that have occured on the uterus. The contraction carrying important information about a birth is more important than other situations (such as baby movement) among all events. Besides, the contraction can occur in the signals includingpregnancy, premature birth or normal birth. In thesecond part, contractions which were analysedby the first classification stage were reconsidered again in terms of understanding the differences of preterm birth, normal birth and pregnancy terms by assigning new artificial neural network units. As a result of this second part of the study preterm birth contractions were isolated from other term contractions successfully. Both of two classification performances were analysed in terms of the use of different bipolar channels. It is reported that the channels givingthe best performance for the certain classifiers foran each classification effort. At the same time, time dependent variations in the features of contractions have been monitored and compared for temporal analysis. In this way, it was aimed to define more distinctive feature associated by time dependent variations.In addition to these studies, an interface forsignal processing and dedicated analysis tools has been developed in the Matlab environment in order to give a better presentation for the results.Through the interface, signal preprocessing studies, energy and correlation analysis, feature extractions and time-frequency analysis can be planned.