Characterization and automatic classification of preterm and term uterine records

Area of Science:

• Biomedical Engineering
• Obstetrics and Gynecology
• Signal Processing

Background:

• Predicting preterm birth remains a significant clinical challenge, driving the search for improved, non-invasive methods.
• Existing research primarily analyzes ElectroHysteroGram (EHG) signals during uterine contractions.

Purpose of the Study:

• To characterize non-contraction (dummy) intervals of uterine EHG and external tocogram (TOCO) signals for the first time.
• To develop and validate a novel method for early preterm birth risk assessment using these signals.
• To establish a new biophysical marker for preliminary preterm birth danger evaluation.

Main Methods:

• Development of a new uterine record dataset (TPEHGT DS) including pregnant and non-pregnant women.
• Quantitative characterization of contraction and dummy intervals using normalized power spectra of EHG and TOCO signals (1.0-2.2 Hz frequency band).
• Application of classification algorithms (sample entropy, median frequency, peak amplitude) to predict preterm birth.

Main Results:

• Peak amplitudes in the 1.0-2.2 Hz band, reflecting maternal heart rate influence, are low in preterm pregnancies and non-pregnant women, unlike term pregnancies.
• Dummy intervals demonstrated comparable or slightly superior classification performance for preterm birth prediction compared to contraction intervals.
• The proposed method achieved 100% accuracy for early records (around 23rd week) and 96.33% overall accuracy (AUC 99.44%) on a public database, outperforming existing methods.

Conclusions:

• The peak amplitude of normalized power spectra in the 1.0-2.2 Hz band serves as a novel biophysical marker for early preterm birth risk assessment.
• Utilizing dummy intervals offers a simple and effective clinical technique for early preterm birth danger assessment, even before contractions are apparent.
• The method's high accuracy and suitability for early pregnancy (from the 23rd week) make it a promising tool for clinical application.