Prediction of Preterm Delivery from Unbalanced EHG Database

Area of Science:

• Biomedical Engineering
• Signal Processing
• Maternal-Fetal Medicine

Background:

• Early prediction of preterm labor is crucial for minimizing complications.
• Electrohysterogram (EHG) signals offer a non-invasive method for monitoring uterine activity.
• Existing methods may require complex signal processing or multiple channels.

Purpose of the Study:

• To develop an automated algorithm for preterm labor prediction using a single EHG signal.
• To evaluate the performance of machine learning classifiers for EHG-based prediction.
• To establish a computationally efficient and accurate method for early preterm labor detection.

Main Methods:

• Empirical Mode Decomposition (EMD) was used to decompose EHG signals into intrinsic mode functions (IMFs).
• Feature extraction included sample entropy (SampEn), root mean square (RMS), and mean Teager-Kaiser energy (MTKE) from each IMF.
• K-nearest neighbors (kNN), support vector machine (SVM), and decision tree (DT) classifiers were employed for prediction.

Main Results:

• The database comprised 262 term and 38 preterm delivery EHG recordings.
• The SVM classifier with a polynomial kernel achieved the highest performance: 99.5% sensitivity, 99.7% specificity, and 99.7% accuracy.
• The Decision Tree (DT) classifier demonstrated strong results with 100% sensitivity, 98.4% specificity, and 98.7% accuracy.

Conclusions:

• The proposed algorithm effectively predicts preterm labor using a single EHG channel.
• This approach surpasses state-of-the-art methods by avoiding synthetic data and feature ranking.
• The single-channel EHG analysis offers a simplified and highly accurate tool for preterm labor prediction.