Artificial intelligence in the respiratory sounds analysis and computer diagnostics of bronchial asthma

Abstract

Using a database containing audio files of respiratory sound records of asthmatic patients and healthy patients, a method of computer-aided diagnostics based on the machine learning technique – creation of neural networks, has been developed. The database contains 952 records of respiratory sounds of asthma patients at different stages of the disease, aged from several months to 47 years, and 167 records of volunteers. Records were carried out with a quiet breathing at four points: in the oral cavity, above the trachea, on the chest, the second intercostal space on the right side, and at a point on the back.The developed method of computer-aided diagnostics allows diagnosing bronchial asthma with high reliability: sensitivity of 89.3%, specificity of 86%, accuracy of about 88% and Youden’s index of 0.753.The program learned once makes it possible to diagnose bronchial asthma with high reliability regardless of patient’s gender and age, a stage of disease, as well as the point of sound recording.The developed method can be used as an additional screening method for the diagnostics of bronchial asthma and serve as the basis for development of computer control methods, including remote control (telemedicine) of patient’s condition and the effectiveness of the applied drugs in real time.

Application of deep learning for bronchial asthma diagnostics using respiratory sound recordings

“…The accuracy of computer-assisted diagnostics of pathological sounds and their classification by type (wheezing, rhonchi, crackles) exceeded the accuracy of the physician’s diagnosis. A computer-based method for diagnosing BA ( Gelman et al, 2021 ) based on machine learning and comparison of respiratory sounds of asthmatic patients and healthy volunteers demonstrates similar values of sensitivity of 89.3%; reliability of 86%, accuracy of about 88%, and Youden’s index of 0.753.…”

“…In pulmonary diseases, pathological processes and the physiological changes provoked by them are reflected in the character of respiratory sounds. The computer-assisted diagnostics methods developed previously that utilize the Fourier spectrum analysis or other sound wave metrics attempt to link disease diagnostics with the analysis of biological, physiological, and clinical processes in the patient’s body ( Furman et al, 2014 ; Furman et al, 2015 ; Furman et al, 2021 ; Furman et al, 2020 ; Gelman et al, 2021 ). Machine learning and deep learning methods are designed to divide data into certain categories according to identified common attributes that cannot always be clearly matched to body processes, and to statistically analyze the signals.…”

Respiratory Diseases Diagnosis Using Audio Analysis and Artificial Intelligence: A Systematic Review

“…However, it has been observed that research conducted in this domain may focus on multiple diseases at the same time if they have similar attributes. Specifically, the authors of [ 43 , 45 ] focus on asthma detection through the identification of wheezes and achieve an accuracy of 75.21% and 88%, respectively, whereas [ 47 ] includes asthma alongside other diseases for disease classification and achieve an accuracy of 60% and 95.67%. One of the most studied diseases for identification or classification is COPD, and [ 46 , 47 ] introduced ML algorithms for COPD identification, while [ 46 ] developed a methodology that reached 100% accuracy.…”

“…Delving into the data processing field that is followed for implementation, the studies [ 43 , 45 , 46 , 47 , 48 , 49 , 51 , 54 , 55 , 60 , 61 , 62 ] perform Feature Extraction on the data; however, the Feature Extraction comes after several preprocessing steps, namely filtering, upsampling or downsampling, or other transformations. The studies [ 45 , 46 , 49 ] extract several statistical and spectral features, such as spectral bandwidth, zero-crossing rate, spectral roll-off, etc. ML algorithms that exploit different statistical and spectral features can see a significant improvement in their prediction capabilities by utilizing feature selection techniques.…”

“…Refs. [ 45 , 55 ] also exploit the RF model, and [ 44 , 56 ] utilizes the K -NN algorithm too. In Ref.…”