ABSTRACT
Respiratory sound samples unveil key information about the patient's lung condition. According to the World Health Organization, respiratory diseases are one of the major causes of death and lung sound analysis is immensely crucial for accurate diagnosis by medical experts. This book chapter presents an automatic lung sound classification approach based on time-frequency image representation and machine learning. The lung sound samples are first converted into spectrogram time-frequency image representations. Textural differences present in abnormal (pneumonia) lung sounds are exploited to classify input lung sound as normal or pathological. Textural clues effectively capture these spectrogram image variations using three different textural features. Finally, decision tree classification is used for labelling the input lung sound sample as healthy or unhealthy. Various performance evaluation parameters are utilised for evaluating the algorithm performance using ICBHI 2017 challenge respiratory sound database. The proposed approach outperforms several existing lung sound classification algorithms in terms of average accuracy.
