A Wheeze Detection Method based on a Time Series Regularity of Time- Frequency Distribution

Abstract

This paper proposes a robust method for wheeze sound detection. The presented approach is based on a time series regularity measure called sample entropy of a time-frequency distribution (Gabor Spectrogram). First, the respiratory sound signals are segmented into their respective inspiration/expiration phases for segment-wise detection of wheeze sounds. Applying Gabor Spectrogram to these extracted segments, timefrequency representation of each segment is obtained. From this representation, regularity of each segment is determined using Sample Entropy. A decision rule is then applied to sample entropy sequences to determine whether wheeze or normal sound. The accuracy of method is tested on wheeze sounds with low and high intensity wheeze inspirations/expirations segments of respiratory sound signals. The experimental results reveal that the overall detection accuracy is 86.25% for inspiration and is 82.5% for expiration.