1. Unsupervised framework for single channel heart and lung sounds separation in data constrained environments.
- Author
-
Ullah, Rizwan and Zhang, Shaohui
- Abstract
• Proposed a cutting-edge technique for cardiopulmonary sound separation. • This paper investigates cardiorespiratory auscultation using uniquely integrated MCRA-OMLSA and NMF-HMM techniques. • Unsupervised learning method to enhance signal accuracy without reliance on external data. Increasing the accuracy and efficiency of cardiopulmonary disease screening is crucial, as cardiovascular illnesses pose a serious threat to human health and result in numerous fatalities annually around the globe. For doctors to assess the illness, auscultation is a non-invasive technique. An electronic stethoscope records heart sounds (HS) and lung sounds (LS), which are useful acoustic data for diagnosing pulmonary diseases. Nevertheless, interference between LS and HS that is visible in the frequency and temporal domains hinders the effectiveness of the diagnostic process. Moreover, conducting any prior training is not always feasible in many real-world situations. It is very challenging to prepare pure lung and heart sounds. This paper proposes an integrated approach using a minima-controlled recursive algorithm (MCRA) equipped with optimally modified-log spectral amplitude (OMLSA), a trainable time wrapping technique (TTW), and a non-negative matrix factorization-hidden Markov model (NMF-HMM) to effectively separate lung and heart sounds, knowing nothing about the heart or lung sounds beforehand. A detailed investigation in terms of unsupervised framework is provided in this paper. This study is the first to apply the integrated OMLSA-TTW-NMF-HMM approach to cardiopulmonary sounds separation and presents the HLS classification method. This research proposes a framework with improved robustness and separation quality compared to existing state-of-the-art techniques. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF