Abstract Odontogenic myxoma (OM) and odontogenic myxofibroma (OMF) are rare benign odontogenic tumors characterized by heterogeneous stromal composition, for which objective and reproducible pathological evaluation remains challenging. This multicenter study aimed to quantitatively assess fibrous tissue proportion (FTP) using AI-assisted digital pathology and to explore its clinicopathological relevance across odontogenic myxoma spectrum lesions. A total of 143 surgical specimens were collected from 34 institutions, and 100 cases were included after centralized pathological review. FTP was independently estimated by two board-certified oral pathologists to generate expert reference data. Whole-slide images of Masson’s trichrome–stained sections were analyzed using a multi-stage deep learning pipeline implemented within a unified digital pathology platform. Agreement between expert assessment and quantitative measurements was evaluated, and associations between FTP and clinical variables were explored. Quantitative FTP measurements showed good agreement with expert pathological evaluation and revealed substantial inter-institutional variability in pathological diagnoses. In clinicopathological analyses, higher FTP was independently associated with unilocular radiological morphology. These findings demonstrate that AI-assisted quantitative pathology provides a reproducible framework for visualizing stromal heterogeneity in odontogenic myxoma spectrum lesions and may support more consistent clinicopathological interpretation across institutions.
Harazono et al. (Fri,) studied this question.