Asthmatic Wheeze Detection From Compressively Sensed Respiratory Sound Spectra.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Respiratory Medicine

Background:

• Wearable sensor systems with smartphones can aid asthma diagnosis and management.
• Compressive sensing (CS) reduces power consumption in wearable sensors by compressing signals.
• Existing methods struggle with wheeze detection accuracy from CS-processed data.

Purpose of the Study:

• To develop a robust algorithm for accurate wheeze detection from compressed respiratory sound data.
• To improve the accuracy of asthma diagnosis using wearable sensor technology.
• To enable efficient data processing for real-time respiratory sound analysis.

Main Methods:

• A novel algorithm utilizing a hidden Markov model (HMM) to identify and track wheeze frequencies.
• Application of the algorithm to compressed short-term Fourier spectra (STFT) reconstructed via orthogonal matching pursuit (OMP).
• Evaluation of algorithm performance on Nyquist-rate sampled respiratory sounds and CS-reconstructed STFT.

Main Results:

• The algorithm achieved 89.34% sensitivity, 96.28% specificity, and 94.91% accuracy on standard STFT.
• Less than 2% loss in classification accuracy was observed when using STFT reconstructed by OMP at a 4x compression ratio.
• The algorithm demonstrated comparable execution speed to existing methods and potential for parallel processing.

Conclusions:

• The developed algorithm enables accurate wheeze detection from significantly compressed respiratory sound data.
• This approach minimizes accuracy loss, making CS-based wearable sensors more viable for asthma monitoring.
• The algorithm offers a promising solution for efficient and accurate real-time respiratory sound analysis in wearable devices.