A multimodal computational pathology framework for differential diagnosis of fibro-osseous jaw lesions
Precision Pathology · 2026
DOI: 10.1016/j.prpath.2026.100012 ↗Kurz gefasst
Fachartikel in der Zeitschrift Precision Pathology, 2026. Verfasst von Hannah Walsh und 5 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: British Society for Oral and Maxillofacial Pathology, Fundação de Amparo à Pesquisa do Estado de São Paulo.
Abstract (Originalsprache)
Fibro-osseous lesions (FOLs) of the jaws show substantial histopathological overlap, complicating diagnosis. This study developed and validated a multimodal artificial intelligence framework integrating stromal nuclear morphometry, clinicopathological variables and deep histopathological features for classifying fibrous dysplasia (FD), cemento-ossifying fibroma (COF) and cemento-osseous dysplasia (COD), and evaluated its performance alongside experienced pathologists. A multicentre retrospective cohort of 157 patients (55 FD, 55 COF and 47 COD) was analysed. Whole-slide images underwent colour normalisation and tissue and nuclear segmentation. Nuclear features were extracted using QuPath, while XGBoost with SHAP was used to evaluate structured variables and provide case-level interpretability. Image embeddings from ResNet-50, DeepLabV3 and InceptionV3 were integrated with clinicopathological and morphometric data. Model performance was assessed using five-fold patient-level cross-validation, internal validation, institution-stratified analysis and independent hold-out validation. Performance was compared with two blinded oral and maxillofacial pathologists. Nuclear area, circularity, eccentricity and mean haematoxylin optical density differed significantly between lesions (all P
Förderung
- British Society for Oral and Maxillofacial Pathology
- Fundação de Amparo à Pesquisa do Estado de São Paulo
Zitieren
APA
Walsh, H., Souza, L. L. D., Bezerra, H. K. F., Lopes, M. A., Vargas, P. A., & Khurram, S. A. (2026). A multimodal computational pathology framework for differential diagnosis of fibro-osseous jaw lesions. Precision Pathology. https://doi.org/10.1016/j.prpath.2026.100012
Vancouver
Walsh H, Souza LLD, Bezerra HKF, Lopes MA, Vargas PA, Khurram SA. A multimodal computational pathology framework for differential diagnosis of fibro-osseous jaw lesions. Precision Pathology. 2026. doi:10.1016/j.prpath.2026.100012
BibTeX
@article{Walsh2026,
author = {Walsh, Hannah and Souza, Lucas Lacerda de and Bezerra, Hélen Kaline Farias and Lopes, Marcio Ajudarte and Vargas, Pablo Agustin and Khurram, Syed Ali},
title = {A multimodal computational pathology framework for differential diagnosis of fibro-osseous jaw lesions},
journal = {Precision Pathology},
year = {2026},
doi = {10.1016/j.prpath.2026.100012}
}
EUVIMED