-Title: PathOsteo: An Interpretable and Uncertainty-Aware Dual-Foundation Framework for Osteosarcoma Histopathology under Patient-Level Validation
-Journal/Conference: 
-Authors: Youngcheol Song, Heebok Kim, and Jongpil Jeong
-DOI: 
-Journal/Conference Link:

Abstract: Background and Objectives: Osteosarcoma is the most common primary malignant bone tumour in children and adolescents [1], and its histopathological diagnosis is limited by inter-observer variability and the scarcity of experienced bone pathologists. Most existing deep-learning studies on osteosarcoma pathology operate as opaque classifiers that report accuracy without exposing diagnostic rationale or predictive uncertainty [3,4]; moreover, they are typically evaluated with tile-level data splitting, which allows tissue from the same patient to appear in both training and validation sets and thereby introduces data leakage. For the task of discriminating viable tumour from non-tumour regions within osteosarcoma tissue, this study proposes PathOsteo, a framework that (i) eliminates data leakage through patient-level cross-validation, (ii) provides interpretable rationale aligned with the World Health Organization (WHO) definition of osteosarcoma [5], and (iii) quantifies predictive uncertainty.

-Status: Submitted (2026/07/02)