A filter bank-based source extraction algorithm for heart sound removal in respiratory sounds

Area of Science:

• Biomedical Engineering
• Signal Processing
• Respiratory Medicine

Background:

• Respiratory sounds (RS) are vital for diagnosing lung conditions.
• Heart sounds (HS) often contaminate RS recordings, hindering accurate analysis.
• Existing methods for HS removal from RS are limited.

Purpose of the Study:

• To propose a novel semi-blind single-channel source extraction algorithm for heart sound removal from respiratory sounds.
• To effectively extract pure respiratory sounds from corrupted signals.
• To evaluate the performance of the proposed method in various respiratory sound recordings.

Main Methods:

• Incorporation of filter banks and template-based matching using Finite Impulse Response (FIR) filters.
• Semi-blind single-channel source extraction algorithm.
• Performance evaluation using average power spectral densities (PSD) comparison over selected frequency bands (20-800Hz).

Main Results:

• The proposed method effectively extracts pure respiratory sounds from heart sound-corrupted signals.
• The average spectral difference between original and reconstructed signals was found to be 2.8707+/-0.9875dB below 800Hz.
• The algorithm demonstrated effectiveness across various respiratory sound recordings.

Conclusions:

• The developed semi-blind source extraction algorithm is effective for heart sound removal from respiratory sounds.
• The method offers a promising approach for improving the quality of respiratory sound recordings for clinical diagnosis.
• Further research can explore real-time implementation and application in diverse clinical scenarios.