Cough sound analysis and objective correlation with spirometry and clinical diagnosis

Abstract

Abstract In India, there are 100 million people who suffer from various respiratory problems; globally it is about 1–1.2 billion. The main problem attributed to the prevalence of respiratory diseases is lack of cost-effective and lab-free methods for early diagnosis. Spirometry is the standard clinical test procedure for detection of respiratory problems, but it requires repetition, and is also expensive and not available in rural areas. Cough sounds carry vital information about the respiratory system and the pathologies involved. Through this study, we detail how a combination of standard signal processing features and domain-specific features play a key role in distinguishing cough patterns. We could establish a relationship between cough pattern and respiratory conditions including widened airway, narrowed airway, fluid filled air sacs, and stiff lungs. Further, cough sound characteristics are correlated to the airflow parameters of spirometry. Our results show strong correlation of cough sound characteristics with airflow characteristics including FEV1, FVC and their ratios, which are important in identifying the type of lung diseases as either obstructive (obstruction in airway) or restrictive (restricts lung expansion). We have constructed a machine learning model to predict obstructive versus restrictive pattern, and validated it using K-fold cross-validation based on ground truth data. With a pattern prediction accuracy of 91.97%, sensitivity of 87.2%, and specificity of 93.69%, our results are encouraging.