A constrained tonal semi-supervised non-negative matrix factorization to classify presence/absence of wheezing in respiratory sounds. (April 2020)

• Record Type: Journal Article
• Authors: Torre-Cruz, J.; Canadas-Quesada, F.; García-Galán, S.; Ruiz-Reyes, N.; Vera-Candeas, P.; Carabias-Orti, J.
• Is Part Of: Applied acoustics. Volume 161(2020)
• Publication Date: 2020-04
• Subjects: Non-negative matrix factorization (NMF) -- Divergence -- Wheezing -- Smoothness -- Monophonic constraint -- Spectral trajectories; Acoustical engineering -- Periodicals; Periodicals; 620.2
• DOI: 10.1016/j.apacoust.2019.107188
• Languages: English
• ISSNs: 0003-682X

Abstract

Highlights: We propose a method to detect the presence of wheeze sounds in breath recordings. Evaluation has been conducted using six datasets of healthy/unhealthy subjects. Our proposal provides promising results compared to the state-of-the-art methods. The proposed method is robust compared to different signal-to-noise ratios. Results show independence from temporal duration of the input breath recording. Abstract: From a clinical point of view, the detection of wheezing presence in respiratory sounds is a challenging task for early identification of pulmonary diseases since wheezing is the main manifestation associated to airway obstruction. In this article, we propose a novel method to detect the presence or absence of wheeze sounds in breath recordings in order to increase the reliability of the subjective diagnosis provided by the physician in the auscultation process. Specifically, it is assumed an unhealthy subject when wheeze sounds can be detected during breathing. The proposed method consists of three stages. The first stage attempts to estimate the spectral interval, band of interest (BOI), that shows the highest probability to find wheeze sounds. In the second stage, a constrained tonal semi-supervised non-negative matrix factorization (NMF) approach is applied to obtain spectral patterns that models the periodic or tonal nature typically shown by wheeze sounds. The third stage analyzes the estimated wheezing spectrogram based on the smoothness of the spectral trajectories from the most significant energy previously factorized in the BOI. Our system has been evaluated and compared to other state-of-the-art methods, yielding competitive results in the wheezing presence detection in respiratory sounds.