Noisy Neonatal Chest Sound Separation for High-Quality Heart and Lung Sounds

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Neonatal Healthcare

Background:

• Remote cardio-respiratory monitoring of neonates relies on high-quality stethoscope-recorded chest sounds.
• Existing methods struggle with noise interference, compromising diagnostic accuracy.

Purpose of the Study:

• To introduce and evaluate novel artificial intelligence-based Non-negative Matrix Factorisation (NMF) and Non-negative Matrix Co-Factorisation (NMCF) methods for neonatal chest sound separation.
• To compare the performance of NMF and NMCF against existing single-channel separation techniques.

Main Methods:

• Generation of an artificial mixture dataset including heart, lung, and noise sounds for quantitative assessment.
• Testing NMF and NMCF on real-world noisy neonatal chest sounds.
• Evaluation using signal-to-noise ratios (SNRs), vital sign estimation error, and a developed signal quality score (1-5).

Main Results:

• NMF and NMCF methods demonstrated superior performance over existing methods, achieving 2.7 dB to 11.6 dB higher SNRs on artificial data.
• Real-world data showed significant signal quality improvement, ranging from 0.40 to 1.12 points.
• Median processing times were 342 ms for NMF and 28.3 s for NMCF per 10s recording.

Conclusions:

• The proposed NMF and NMCF methods offer stable and robust performance for denoising neonatal heart and lung sounds.
• These AI-driven techniques are valuable for improving remote cardio-respiratory health monitoring in neonates.
• The methods enhance the reliability of analyzing chest sounds in real-world clinical environments.