Hidden Markov model in nonnegative matrix factorization for fetal heart rate estimation using physiological priors

Area of Science:

• Biomedical Engineering
• Maternal-Fetal Medicine
• Signal Processing

Background:

• Fetal heart rate (fHR) analysis is crucial for monitoring fetal well-being during labor.
• Cardiotocography (CTG) is the standard, but has limitations.
• Abdominal ECG (aECG) offers a promising alternative for fHR measurement.

Purpose of the Study:

• To improve fetal heart rate estimation from abdominal ECG signals.
• To address the limitations of existing Non-negative Matrix Factorization (NMF) methods by incorporating temporal dynamics.
• To enhance the accuracy and robustness of fHR analysis for detecting fetal distress.

Main Methods:

• Developed an enhanced fHR estimation method combining Non-negative Matrix Factorization (NMF) with Hidden Markov Models (HMM).
• Integrated physiological information on fHR temporal evolution into the NMF framework using Bayesian priors.
• Evaluated the proposed method on 23 real-world aECG signals against CTG reference.

Main Results:

• The proposed NMF-HMM method demonstrated improved performance compared to the NMF-only algorithm.
• Agreement between the new method and CTG reference increased from 71% to 80%.
• The enhanced modelization of fHR characteristics led to more robust estimation.

Conclusions:

• Combining NMF with HMM significantly enhances fHR estimation accuracy from aECG.
• The improved accuracy suggests a more reliable tool for monitoring fetal well-being during labor.
• This approach highlights the importance of incorporating temporal dynamics for robust fetal monitoring.