Spectral Correlation Density based Electrohysterography Signal Analysis for the Detection of Preterm Birth

Abstract

Preterm birth (gestational age <37 weeks) is one of the most critical global concerns that causes maternal and fetal morbidity and mortality. Early detection of this condition allows for timely intervention to delay labor by providing tocolytic drugs and rest. The objective of this work is to explore the cyclostationary behavior in electrohysterography (EHG) signals and to predict preterm conditions. The signals recorded prior to the 26 weeks of pregnancy are considered in this work. It is pre-processed using Butterworth bandpass filters to remove artifacts. The fast Fourier transform accumulation method (FAM) is applied to the pre-processed signals to estimate the spectral correlation density (SCD). The degree of cyclostationarity (DCS) is calculated from SCD to evaluate the presence of cyclostationarity in the signals. Features, such as mean, variance, cyclic frequency spectral area (CFSA), and full width half maximum (FWHM), are extracted from the spectra and statistically analyzed. The results illustrate that SCD and DCS confirm the existence of cyclostationarity in EHG signals. All the extracted features are observed to decrease in preterm conditions. This might be due to the increased coordination that is reflected in the signal in terms of reduced frequency components. Further, extracted features are found to have statistical significance (p < 0.05) in discriminating both the conditions. Thus, it appears that cyclostationary features might be clinically beneficial in the early prediction of preterm birth.