Research Projects

Respiratory diseases, including chronic obstructive pulmonary disease (COPD), tuberculosis (TB), and asthma, pose significant global health challenges, with a disproportionately high burden in resource-limited settings such as India. These conditions are exacerbated by factors such as rising air pollution and limited healthcare infrastructure, creating an urgent need for accessible and scalable diagnostic solutions. Traditional diagnostic methods, including lung auscultation, chest X-rays, and molecular tests, face limitations related to subjectivity, cost, and infrastructure requirements, making them unsuitable for large-scale deployment in underserved regions. Emerging technologies, particularly artificial intelligence (AI) and digital sound analysis, offer transformative potential for overcoming these barriers. Respiratory sounds—comprising cough, breathing, and speech—carry diagnostically relevant information that can be analyzed using machine learning models to detect subtle acoustic patterns associated with specific diseases. Building on promising preliminary results, including AI-based COVID-19 screening tools developed by our team, this project seeks to improvize respiratory diagnostics by integrating AI-driven acoustic analysis with widely available smartphone technology. The primary objectives of this study are: (1) to create a comprehensive, medically validated respiratory audio dataset from diverse Indian populations; (2) to develop advanced AI models, including a respiratory audio foundation model capable of extracting robust disease-specific features; and (3) to design and validate a user-friendly, smartphone-based diagnostic tool for point-of-care testing. The dataset will include recordings of breathing, cough, and speech sounds from healthy individuals and patients with conditions such as asthma, COPD, TB, and pneumonia. The project will develop models capable of capturing diagnostically significant acoustic features. A foundation model will be built using advanced machine learning architectures such as transformers and self-supervised learning frameworks, enabling generalizability across diseases and populations. This model will generate compact embeddings that capture the essence of respiratory audio, allowing fine-tuning for specific conditions and enhancing performance in data-constrained scenarios. Model transparency will be ensured through the integration of explainable AI (XAI) techniques, which will elucidate the basis of predictions, fostering trust and clinical adoption. The final deliverable will be a smartphone-based point-of-care diagnostic tool that integrates these AI models. Designed with a user-friendly interface, the tool will enable real-time analysis of respiratory sounds, providing accurate, non-invasive diagnostics to healthcare providers and patients in resource-limited settings. Its performance will be validated through clinical studies.