Clinical validation of Swaasa® artificial intelligence platform using cough sounds for screening and diagnosis of respiratory diseases

<bold>Introduction:</bold> Analysis of cough sounds has the potential to give a clue to the underlying respiratory disease. The Swaasa® platform using artificial intelligence technology was evaluated to diagnose respiratory diseases. <bold>Methods:</bold> Patients (age >= 18 years) with respiratory symptoms were recruited into the study. Patient’s cough sound was recorded by a mobile device with Swaasa software. Spirometry was done and clinical diagnosis was taken as reference standard. Variables were analysed in binary form as normal and abnormal. Multi-class outcomes were defined for spirometry patterns classified as obstructive, restrictive or normal, for severity of impairment and for disease conditions such as asthma, COPD, ILD, bronchiectasis etc. <bold>Results:</bold> We recruited 2179 patients with respiratory diseases and 827 normal individuals. 62% of them were males. The Swaasa platform showed high accuracy in distinguishing normal individuals from those with disease as well as distinguishing normal from abnormal spirometry (Fig 1). The performance was consistent across demographic parameters such as age, gender, and BMI. Its accuracy in differentiating into various patterns of spirometry abnormality (obstructive/restrictive), severity of spirometry abnormalities as well as disease types is sub-optimal. (Fig 2) <bold>Conclusion:</bold> Swaasa® commends a high level of precision in classifying normal and abnormal conditions across demographic subgroups, indicating its promise as a screening tool. Performance for specific respiratory conditions and disease severity needs further refinement to improve it accuracy.