Windowed multivariate autoregressive model improving classification of labor vs. pregnancy contractions

Area of Science:

• Biomedical Engineering
• Signal Processing
• Obstetrics

Background:

• Analyzing uterine electrical activity is crucial for labor detection and preterm labor prediction.
• Existing methods for analyzing uterine signals face challenges with non-stationary data.
• Multivariate autoregressive (MVAR) modeling is a recent promising approach.

Purpose of the Study:

• To introduce a time-varying version of the MVAR model, termed W-MVAR.
• To investigate the connectivity of uterine electrical signals while accounting for non-stationarity.
• To evaluate the W-MVAR method's efficacy in distinguishing labor and non-labor states.

Main Methods:

• Development of the windowed multivariate autoregressive (W-MVAR) model.
• Testing the W-MVAR model on synthetic, non-stationary uterine signal data.
• Application of the W-MVAR model to real, multi-channel uterine electrograms.

Main Results:

• W-MVAR demonstrated superior performance over standard MVAR in detecting non-stationary connectivity in synthetic data.
• The W-MVAR model effectively differentiated between non-labor and labor signals in real uterine recordings.
• Analysis revealed the W-MVAR method's capability to capture dynamic changes in uterine signal patterns.

Conclusions:

• The proposed W-MVAR method is a robust tool for analyzing non-stationary uterine electrical activity.
• W-MVAR shows significant potential for clinical applications in accurate labor detection.
• This technique offers a promising advancement for the prediction of preterm labor.