Recognition of premature births by bispectrum-based abdominal electromyography signal processing

A novel technique for detection and recognition of normal and premature births is proposed and experimentally examined. It is based on the extraction of novel class of informative features contained in higher-order spectrum, namely, bispectrum of the abdominal electromyography signals registered on the abdominal surface of pregnant woman. It is demonstrated that the amplitude bispectrum, phase bispectrum and bicoherence signatures computed for electromyography signals can serve as the perspective facilities for detection and recognition of the normal and premature births. The proposed bispectrum-based information features were studied by real-life experimental data processing. Uterine activity corresponding to two weeks and one week before birth for several patients has been investigated. Experimental results obtained for a number of the patients demonstrate the possibility to extract novel classification features contained in the computed biamplitude, biphase and bicoherence signatures.