Abstract Background Granulomatous inflammation is a key diagnostic feature in diffuse lung diseases (DLDs) such as hypersensitivity pneumonitis (HP), sarcoidosis, and granulomatosis with polyangiitis (GPA). However, interpretation of granuloma morphology varies across pathologists, which may contribute to diagnostic inconsistency in multidisciplinary ILD evaluation. We sought to establish reproducible morphological criteria through a structured expert consensus process, aiming to standardize granuloma classification and support clinical decision-making in DLDs. Methods Ninety multidisciplinary-diagnosed DLD cases yielded 863 granulomatous lesions for review. Thirteen pulmonary pathologists classified granuloma subtypes in two rounds. After initial independent review, pathologists reassessed all images following group feedback (Delphi method). Interobserver agreement and diagnostic behavior patterns were analyzed. Images reaching ≥70% consensus in the most reproducible cluster were included in a standardized atlas. Results Consensus review significantly improved agreement among experts (Fleiss’ κ 0.27→0.47, p 0.001). Delphi analysis identified two diagnostic clusters; one demonstrated greater internal consistency (κ 0.52) and stronger ability to distinguish DLD etiologies (AUC 0.797). Consensus from this cluster formed the basis for a 485-image reference atlas. Distinct granuloma patterns were confirmed across diseases: sarcoidosis and GPA predominantly showed well-formed/palisading granulomas, while HP and IPF were characterized mainly by poorly formed granulomas, although occasional well-formed granulomas were consistently identified in HP/IPF as well. Conclusion A structured Delphi approach improved diagnostic reproducibility in granuloma assessment and identified a high-performance expert cluster whose consensus provides reliable criteria for clinical application. The resulting standardized atlas offers a practical tool for harmonizing granuloma interpretation in multidisciplinary ILD care and lays a foundation for future training and machine-learning integration in granulomatous lung disease diagnostics. This abstract is funded by: nothing
Sato et al. (Fri,) studied this question.