Comprehensive application of artificial intelligence in preserved ratio impaired spirometry: A systematic literature review.

Abstract

Artificial intelligence (AI) has expanded into respiratory disease diagnosis, subtyping, and prognosis, enabling early detection and precision care. However, AI applications in Preserved Ratio Impaired Spirometry (PRISm) remain nascent. This study analyzes this gap to guide future research. A systematic review was conducted to analyze the application of AI in PRISm, searching across PubMed, Cochrane Library, Web of Science, Ovid Medline, Scopus and Embase. A total of eleven studies were included, all of which focused on diagnostic and classification tasks. Among these, three utilized radiomics models, four employed machine learning algorithms, two integrated machine learning with radiomics, and two applied deep learning approaches. Nine studies were published within the past two years, with results demonstrating the high performance and developmental potential of AI technologies in this domain. AI research on PRISm spans multiple disciplines, including exhaled metabolomics, environmental exposure assessment, radiomics, and deep learning. Existing studies have preliminarily validated the technical feasibility of artificial intelligence for the early identification of PRISm from multiple perspectives, including imaging, metabolism, and environmental exposure. Modeling strategies that integrate multi-source data have demonstrated superior discriminatory performance compared to single-modality approaches. In the future, with the integration and sharing of multicenter data under privacy-compliant conditions, coupled with the continuous evolution of algorithm architectures toward enhanced generalizability, AI applications for PRISm are expected to transition from static identification to dynamic early warning, thereby providing more robust technical support for precise risk stratification of this condition.