Automated Diagnosis of Pulmonary Diseases Using Lung Sound Signals

AbstractPulmonary diseases are the third biggest cause of deaths worldwide. A prominent method to detect these diseases is the observation of lung sound signals. There is an increasing need for an efficient technique that can automatically diagnose such diseases with high accuracy. In this paper, two popular public datasets are considered, and every lung sound signal is decomposed into 8 frequency bands using rectangular zero-phase filters. Features are extracted from every band, including energy, kurtosis, mean absolute deviation and Lp norm. The extracted features are utilized for classification using machine learning schemes. The proposed method achieves 99.9% accuracy for multi-class classification on the combined dataset, and for binary classification, we have evaluated normal signal versus pathogenic signal, which is found to be 100% accurate for most of the diseases. High accuracy is obtained for the individual datasets as well. Top 20 features selected using minimum redundancy maximum relevance algorithm also yield 98.6% accuracy. Therefore, the proposed method can be easily deployed in real-time systems.KEYWORDS: Feature rankingfrequency bandsminimum redundancy maximum relevancepulmonary diseasessignal decompositionzero-phase filters Disclosure statementNo potential conflict of interest was reported by the author(s).Additional informationNotes on contributorsU. HassanU Hassan received her BTech degree in electronics and communication engineering from Kurukshetra university, Haryana, India in 2016 and MTech degree in electronics and communication engineering from Kurukshetra university, Haryana, India. She is currently pursuing PhD at the department of electronics and communication engineering, NSUT, Delhi, India. Her areas of interest include medical signal processing, speech signal processing, artificial intelligence, and machine learning. Email: umaisathakur@gmail.comA. SinghalA Singhal completed his dual degree with BTech in electrical engineering and MTech in information and communication technology from IIT Delhi in 2009. He completed his PhD in molecular communication from IIT Delhi in 2016. He has a total teaching experience of more than 13 years. He is currently working as assistant professor in the department of electronics and communication engineering in NSUT, Delhi, India. His areas of interest are Image Processing, Molecular Communications, Next generating communication technologies, Image retrieval, Theory and Applications of Fourier methods. Corresponding author. Email: amit@nsut.ac.in