Optimal Electrohysterography Signal Preprocessing for Delivery Term Prediction Using Hypergraph Neural Networks.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Accurate prediction of infant delivery term is vital for managing premature births and improving neonatal outcomes.
• Electrohysterogram (EHG) signals offer a non-invasive method for monitoring uterine activity.
• Current preprocessing techniques for EHG signals require optimization for robust delivery term prediction.

Purpose of the Study:

• To identify the optimal preprocessing scheme for electrohysterogram (EHG) signals used in classifying uterine contractions.
• To evaluate the impact of various EHG denoising techniques on the accuracy of delivery term prediction.
• To establish a robust methodology for improving the prediction of infant delivery using AI and EHG data.

Main Methods:

• Utilized a hypergraph neural network (HGNN) to assess different EHG signal preprocessing strategies.
• Investigated signal segmentation, concatenation, standardization, and normalization techniques.
• Compared the performance of the optimized preprocessing pipeline against conventional EHG denoising methods like CCA and EMD.

Main Results:

• The optimal preprocessing involved segmenting, concatenating, standardizing/normalizing, and re-segmenting EHG signals.
• This optimized methodology achieved a prediction accuracy of 89.2% using the HGNN classifier.
• Conventional denoising methods (CCA, EMD, EMD-CCA) did not yield significant improvements in prediction accuracy.

Conclusions:

• Effective preprocessing of EHG signals is critical for enhancing the accuracy of delivery term prediction.
• The proposed preprocessing pipeline offers a significant improvement over existing methods for EHG signal analysis.
• This approach holds clinical relevance for early intervention in potential premature deliveries.