Spirometry Parameter Prediction Using Acoustic Characteristics of Cough

Background: Spirometry evaluates lung function by measuring airflow postmaximal inspiration, using parameters such as forced expiratory volume in the first second (FEV1), forced vital capacity (FVC), and the FEV1/FVC ratio for respiratory disease classification and severity monitoring. Cough, by inducing intrathoracic pressure changes, reflects respiratory pathology, impacting airflow velocities and cough sound properties. A correlation exists between spirometry values representing air flow-volume properties and acoustic features of cough sounds. Aims and Objectives: Using machine learning models trained on cough sound data annotated with corresponding spirometry outcomes, this research demonstrates the ability of Swaasa to predict spirometry values. Materials and Methods: The present study explores the correlation between cough sounds and spirometry values (FEV1, FVC, and FEV1/FVC ratio) for assessing respiratory health. Spirometry parameter regression models were evaluated on 2506 records collected from 1202 subjects. Results: The regression algorithm for predicting spirometry parameters yields optimal results with FEV1/FVC prediction showing 70.47% accuracy, 77.37% sensitivity, and 68.54% specificity. FVC prediction demonstrates 66.04% accuracy, 87% sensitivity, and 48.38% specificity. Conclusion: The study underscores the potential of artificial intelligence-based cough sound analysis for detecting respiratory abnormalities, offering a promising avenue for diagnosis in resource-limited settings.