Non-invasive electrohysterogram (EHG) could be a promising technique for the preterm birth prediction, which could enable us to diagnose the preterm birth before the labor and reduces the infant mortality and morbidity. Previous studies on the preterm birth prediction with EHG have conducted comprehensive researches on various signal features and classification algorithms, but most of them adopted prefilters based on the linear transforms using fixed basis function, although they are suboptimal for the nonlinearity and nonstationarity of the EHG signal. In this paper, multivariate empirical mode decomposition (MEMD) is applied to decompose the electrical activity signal measured on the uterus. After the decomposition, features are calculated for the corresponding oscillations to the uterine contraction. To investigate the performance of the features, three-channel EHG signals of 254 patients (224 term, 30 preterm) are chosen among 300 patients from Physionet term-preterm electrohysterogram (TPEHG) database to extract features from the EHG signals and classify the features using machine learning algorithms. Classification results shows that the proposed method with MEMD achieved 94.66% correctly classified rate (CCR) and 0.987 area under the curve (AUC), which outperformed those with IIR filter implying MEMD provides a new prospect to improve the current preterm birth prediction approach.

The Strong Uncorrelating Transform Complex Common Spatial Patterns (SUTCCSP) algorithm, designed for multichannel data analysis, has a limitation on keeping the correlation information among channels during the simultaneous diagonalization process of the covariance and pseudocovariance matrices. This paper focuses on the importance of preserving the correlation information among multichannel data and proposes the correlation assisted SUTCCSP (CASUT) algorithm to address this issue. The performance of the proposed algorithm was demonstrated by classifying the motor imagery electroencephalogram (EEG) dataset. The features were first extracted using CSP algorithms including the proposed method, and then the random forest classifier was utilized for the classification. Experiments using CASUT yielded an average classification accuracy of 78.10 (%), which significantly outperformed those of original CSP, Complex Common Spatial Patterns (CCSP), and SUTCCSP with p-values less than 0.01, tested by the Wilcoxon signed rank test.