Artificial Intelligence in Respiratory Medicine: Applications, Methodological Challenges, and Clinical Translation

Abstract

Artificial intelligence (AI) is rapidly transforming respiratory medicine by enabling the integration and analysis of complex multimodal data, including imaging, physiological signals, electronic health records, and patient-generated information. This invited state-of-the-art review synthesizes current evidence on AI applications across chronic airway diseases, lung imaging, thoracic surgery and oncology, and sleep medicine and non-invasive ventilation, with a strong focus on clinically meaningful use cases. We highlight how AI can support earlier and more accurate diagnosis, refined phenotyping and endotyping, improved risk stratification, and more individualized therapeutic strategies. Representative applications include AI-assisted spirometry interpretation, multimodal exacerbation prediction, advanced quantitative imaging, perioperative risk modelling, wearable-based sleep apnoea detection, and AI-assisted respiratory support decision-making. Alongside these opportunities, we discuss cross-cutting challenges that currently limit routine implementation, including data quality and representativeness, insufficient external validation and generalizability, algorithmic bias, limited explainability of complex models, and concerns regarding privacy and data governance. Throughout the review, we argue that AI should be conceived as a tool to augment, rather than replace clinical judgment, embedded within human-in-the-loop frameworks that preserve transparency, fairness, and accountability. We conclude by outlining priorities for future research and policy, such as prospective clinical evaluation, multicenter validation in diverse populations, and closer collaboration between clinicians, data scientists, and regulators, which will be essential to translate the promise of AI into tangible improvements in respiratory outcomes and patient-centered care.

Artificial intelligence (AI) is rapidly transforming respiratory medicine by enabling the integration and analysis of complex multimodal data, including imaging, physiological signals, electronic health records, and patient-generated information. This invited state-of-the-art review synthesizes current evidence on AI applications across chronic airway diseases, lung imaging, thoracic surgery and oncology, and sleep medicine and non-invasive ventilation, with a strong focus on clinically meaningful use cases. We highlight how AI can support earlier and more accurate diagnosis, refined phenotyping and endotyping, improved risk stratification, and more individualized therapeutic strategies. Representative applications include AI-assisted spirometry interpretation, multimodal exacerbation prediction, advanced quantitative imaging, perioperative risk modelling, wearable-based sleep apnoea detection, and AI-assisted respiratory support decision-making. Alongside these opportunities, we discuss cross-cutting challenges that currently limit routine implementation, including data quality and representativeness, insufficient external validation and generalizability, algorithmic bias, limited explainability of complex models, and concerns regarding privacy and data governance. Throughout the review, we argue that AI should be conceived as a tool to augment, rather than replace clinical judgment, embedded within human-in-the-loop frameworks that preserve transparency, fairness, and accountability. We conclude by outlining priorities for future research and policy, such as prospective clinical evaluation, multicenter validation in diverse populations, and closer collaboration between clinicians, data scientists, and regulators, which will be essential to translate the promise of AI into tangible improvements in respiratory outcomes and patient-centered care.