AI model improves identification of Ewing sarcoma in biopsy samples
Researchers from the Universitat Politècnica de València and the University of Valencia developed an AI system to tell Ewing sarcoma apart from several look-alike tumors. The model reached 91.6% accuracy and was designed to preserve limited biopsy tissue while guiding additional tests. The study, published in the International Journal of Molecular Sciences, aims to support diagnosis of rare tumors in children and young adults.
Researchers trained and tested the system on 1,926 digitized tissue samples from 729 patients across four Spanish and Italian hospitals. The dataset included 517 Ewing sarcoma cases and several mimicking tumor types, such as rhabdomyosarcoma, chondrosarcoma, GIST, and synovial sarcoma.
The model’s overall accuracy was 91.6%, with 97.1% sensitivity for Ewing sarcoma. Its error rate distinguishing Ewing sarcoma from rhabdomyosarcoma was 1.96%. The work, published in the International Journal of Molecular Sciences, was led by UPV’s CVBLab and Artikode Intelligence with University of Valencia pathologists.
For children and young adults with suspected rare tumors, this tool could help pathologists make better use of tiny biopsy samples and choose follow-up tests more efficiently. That may reduce diagnostic uncertainty and tissue waste, though clinical benefit would depend on validation, regulation, and integration into pathology workflows. Families and clinicians facing difficult look-alike diagnoses could gain additional decision support, but the model is not described as replacing expert review.