Detection of preterm birth in electrohysterogram signals based on wavelet transform and stacked sparse autoencoder

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Preterm birth is a significant global health concern.
• Accurate and early detection of preterm birth is crucial for improving neonatal outcomes.
• Electrohysterogram (EHG) signals offer a non-invasive method for monitoring uterine activity.

Purpose of the Study:

• To develop and evaluate a novel method for preterm birth detection using electrohysterogram (EHG) signals.
• To leverage wavelet-based nonlinear features and a stacked sparse autoencoder (SSAE) for enhanced classification accuracy.
• To compare the performance of the proposed SSAE-based classifier with traditional machine learning models.

Main Methods:

• EHG signals were processed using three-level wavelet decomposition.
• Nonlinear features, including sample entropy, were extracted from approximation and detail coefficients.
• A stacked sparse autoencoder (SSAE) was implemented as the primary classifier.
• The SSAE classifier was benchmarked against Extreme Learning Machine (ELM) and Support Vector Machine (SVM).

Main Results:

• The SSAE-based classifier achieved a high accuracy of 90%, sensitivity of 92%, and specificity of 88%.
• The proposed method demonstrated superior classification performance compared to ELM and SVM.
• Wavelet-based nonlinear features proved effective in capturing relevant EHG signal characteristics.

Conclusions:

• The developed wavelet-based nonlinear feature extraction combined with SSAE offers an effective approach for preterm birth detection.
• The proposed framework exhibits higher discriminability and improved performance over existing techniques.
• This method holds promise for clinical application in early identification of preterm birth risk.