Screening COVID-19 by Swaasa AI platform using cough sounds: a cross-sectional study

The advent of artificial intelligence (AI) has enabled the use of auditory data for disease detection, including COVID-19. Given the global impact of SARS-CoV-2 and the limitations of existing rapid screening methods, this study developed and validated the Swaasa AI platform, a Software as a Medical Device that analyzes 10-second cough recordings and symptoms to screen and prioritize potential COVID-19 cases. Using cough data collected in clinical settings, two parallel models—a CNN on MFCC spectrograms and a feedforward ANN on tabular features—were trained, and their final outputs combined to predict disease likelihood. In clinical validation on 234 suspects, the model achieved 75.54% accuracy, 95.45% sensitivity, and 73.46% specificity. In a pilot test on 183 presumptive cases, 58 of 82 model-positive subjects were truly COVID-19 positive (PPV 70.73%). The approach offers a cost-effective, noninvasive, and scalable screening tool. Further validation in larger and more diverse populations is warranted to improve global applicability.