Development of Artificial Intelligence for Quantitative Assessment of Nasal Inflammatory Cytology in Chronic Rhinitis by Whole‐Slide Images

BACKGROUND: Chronic rhinitis (CR) is currently recognized as a syndrome that manifests in different phenotypes. We aimed to establish an artificial intelligence system (quantitative assessment of nasal inflammatory cytology, QANIC) on the basis of whole-slide images (WSIs) to enable quantitative assessment of nasal inflammatory cells. METHODS: During the development phase of QANIC, we screened nasal secretion smears from 145 CR patients for deep learning and obtaining a robust model. Subsequently, QANIC was applied to an internal cohort (N = 881) and an independent external validation cohort comprising two clinical centers (N = 234). Cluster analysis was employed to analyze two inflammatory variables (nasal and blood eosinophil [Eos] percentages) to investigate the clinical characteristics and inflammatory patterns of different clusters. RESULTS: Three clusters of inflammatory phenotypes were defined in CR patients: Cluster 1 (high nasal and high blood Eoss, accounted for 17.14% and 16.24% in the two cohorts, respectively), Cluster 2 (high nasal but low blood Eoss, 45.86% and 45.30%), and Cluster 3 (low nasal and low blood Eoss, 37.00% and 38.46%). Compared to Cluster 3, Clusters 1 and 2 demonstrated more severe clinical symptoms and nasal Type 2 inflammation, along with a diagnostic advantage in identifying seasonal allergic rhinitis. CONCLUSIONS: The QANIC marks the first time deep learning has been combined with WSIs for nasal cytology diagnosis. Subtyping rhinitis patients based on nasal cytology play an important role in monitoring inflammation dynamics and individualizing treatment.