26th IEEE Signal Processing and Communications Applications Conference (SIU), İzmir, Turkey, 2 - 05 May 2018
The aim of this study is to detect the events that occur on uterus EMG signals and classify related contraction events for preterm birth prediction. In addition to preterm birth, normal birth and pregnancy terms one can find respiratory of mother and baby, muscle movements caused of baby, noisy signals caused of electrodes while recording the signals. Only the contractions from these events that occur on the signal give information about premature births, normal births or pregnancy signals are required to be isolated from other events. Signals obtained from databases in accordance with this purpose are analysed for identification and segmentation by using Wavelet transformation and Teager Energy Operator systematically. As a result of this analysis 491 events are detected. In this phase, within the framework under which the results are evaluated and labeled by experts contraction lookalike segments are determined by using neural networks and classification is achieved with %79.4 accuracy and with %86.4 performance ratio for contraction type detections within the isolated events found in the previous state. Preterm birth contractions, normal birth contractions, pregnancy contraction differences were observed using different time records of the same individual and preterm birth contraction signals were classified with an overall success rate of 82.6% by using artificial neural networks.