Systematic Review on the Ethics of Collecting and Using Respiratory Sound Datasets for the Development of AI-Based Diagnostic Models in the Biomedical Field

Abstract

The rapid advancement of artificial intelligence (AI) for respiratory disease diagnostics has intensified reliance on large-scale respiratory sound datasets, raising complex ethical challenges related to privacy, consent, ownership, and data governance. This systematic review examines the ethical integrity of studies involving cough, breath, and lung sound datasets used for AI-based biomedical applications between 2015 and 2025. Using the PRISMA 2020 framework, 52 eligible studies were identified across major academic databases and evaluated through multidimensional ethical criteria, including transparency of consent processes, adequacy of anonymization, governance mechanisms, dataset licensing, and bias mitigation. The findings reveal significant ethical inconsistencies: less than half of the studies reported clear consent procedures; anonymization techniques were largely insufficient due to the biometric nature of respiratory acoustics; and dataset licensing commonly lacked clarity regarding commercial use. Substantial demographic and clinical biases were also observed, posing risks of inequitable diagnostic performance across population subgroups. The review concludes that current practices exhibit a structural gap between technological innovation and ethical maturity, necessitating stronger governance, standardized licensing, dynamic consent models, and traceable data provenance. Strengthening ethical infrastructures is essential to ensure that AI-enabled respiratory diagnostics advance in a manner that upholds participant rights, clinical safety, and public trust.