Manuscript received May 21, 2024; revised June 10, 2024; accepted July 29, 2024; published January 27, 2025.

Abstract—Respiratory diseases or lung diseases such as asthma bronchiectasis cystic fibrosis are a serious disease. Approximately 8 million people died in each year by chronic obstructive pulmonary disease, lower respiratory tract infections, trachea, bronchial and lung tumors. In addition, COVID-19 is prevalent worldwide in recent years. To analyze these symptom, auscultation of respiratory sounds is very important for screening the respiratory disease. However, there is no quantitative evaluation method for the diagnosis of respiratory sounds until now. To overcome this problem, it is necessary to develop a system to support the diagnosis of respiratory sounds. In the development of support system for auscultation, research by a large-scale, open database used in ICBHI (The International Conference on Biomedical and Health Informatics) 2017 Challenge is in progress. It is expected that a general purpose and highly accurate system will be developed using this dataset. We describe an algorithm for the automatic classification of the respiratory sounds as crackles, wheeze, both, and normal. We improve the classification rates compared with other ICBHI 2017 Challenge teams based on three components. First, we generate the spectrogram images by short-time Fourier transformation. We also extract features using a convolutional recurrent neural network. Third, we classify unknown respiratory sounds by bagging k-nearest neighbor algorithm. In the experiment, we applied our proposed method to 920 respiratory sound data which is obtained by the ICBHI Challenge data sets, and achieved Sensitivity with 0.670, Specificity with 0.863, ICBHI Score with 0.766 respectively. Also, area under the curve based on receiver operating characteristic curve of normal class with 0.892, crackle with 0.891, wheeze with 0.874, both with 0.883 were obtained respectively.

Keywords—respiratory sounds classification, computer aided diagnosis, short-time Fourier transform, convolutional recurrent neural network, k-nearest neighbor algorithm

Cite: Koki Minami, Huimin Lu, Tohru Kamiya, and Shoji Kido, "Automatic Classification of Respiratory Sounds Based on Convolutional Recurrent Neural Network and Bagging k-Nearest Neighbor," Journal of Image and Graphics, Vol. 13, No. 1, pp. 46-51, 2025.

Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC-BY-4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.