Adaptive Filtering for the Maternal Respiration Signal Attenuation in the Uterine Electromyogram

Area of Science:

• Biomedical Engineering
• Signal Processing
• Obstetrics

Background:

• The electrohysterogram (EHG) records uterine electrical activity but is often contaminated by maternal respiration electromyographic signals (MR-EMG).
• Alvarez (Alv) waves within the EHG are significant for predicting preterm and term birth, making MR-EMG interference a critical issue for pregnancy monitoring.
• Effective denoising is essential to isolate Alv waves and improve the accuracy of predictive models for birth outcomes.

Purpose of the Study:

• To develop and validate an algorithm for automatic selection of optimal adaptive filters and their parameters to denoise EHG signals.
• To specifically attenuate MR-EMG interference from Alv waves without distorting the underlying signal.
• To enhance the utility of Alv waves for applications such as preterm birth prediction.

Main Methods:

• An algorithm was created to automatically select the best adaptive filter and parameters from a pool of sixteen candidates using synthetic data.
• The algorithm evaluated filters including Wiener, recursive least squares (RLS), householder recursive least squares (HRLS), and QR-decomposition recursive least squares (QRD-RLS).
• Optimized parameters identified were filter length (L=2) and forgetting factor (λ=1) for specific filters, which were then applied to real EHG data.

Main Results:

• The Wiener, RLS, HRLS, and QRD-RLS filters emerged as top performers after optimization.
• Application of the optimized filters to real data demonstrated significant attenuation of MR-EMG interference within Alv waves.
• The Wiener filter achieved power reductions of -16.74% (Q1), -20.32% (median), and -15.78% (Q3) for MR-EMG in Alv waves (p < 1.31 × 10^-12).

Conclusions:

• The developed automatic adaptive filter selection algorithm effectively reduces MR-EMG interference in EHG signals.
• This method preserves the integrity of Alv waves, thereby improving their reliability for clinical applications like preterm birth prediction.
• The algorithm's optimization approach shows promise for broader applications in signal processing and noise reduction.