Uterine electromyography signal feature extraction and classification

Abstract

Uterine electromyography (EMG) signal has potential for early diagnosis of preterm labour clinically. But it is difficult to differentiate the EMG signal patterns leading to preterm birth. In this paper, the effort has been made to find effective algorithms for extracting the features from uterine EMG signals which can be used to classify the normal term labour from abnormal preterm labour signals. A combined algorithm has been proposed, in which the signal is firstly pre-processed to eliminate the noise and high frequency components. Then, the fractal dimension value along the signal is calculated to identify the abnormal values for distinguishing contraction patterns. Two techniques are employed: phase space reconstruction and singular value decomposition. Finally, the signals are classified using artificial neural network method. The experiment tests indicate that the method can be a choice for solving the problem described but more tests are required to draw final conclusion.