Researchers have introduced Bone Tumor Imaging Reporting and Data System 2.0 (BTI-RADS 2.0), a structured reporting system that could significantly improve how clinicians stratify and manage patients with solitary bone tumors.
In a large multicenter study spanning over 10 years and 10 institutions, the researchers found that a machine learning algorithm trained on structured radiologic reports was capable of distinguishing between benign and malignant bone lesions with high accuracy—achieving an F1 score of 0.81, just shy of the 0.83 score demonstrated by 28 experienced radiologists.
"Standardized bone tumor reporting is crucial for consistent, risk-aligned patient management. Current systems are based on expert consensus and/or lack multicenter validation," noted lead study author Astrée Lemore, of the CHRU de Nancy Pôle Imagerie at the Service d’imagerie Guilloz in France, and colleagues. The machine learning model, based on an XGBoost classifier trained on 27 selected features, represented a major step toward objective, reproducible assessment.
Beyond diagnosis, the team proposed the BTI-RADS 2.0—a new framework that categorizes model outputs into clinically meaningful risk classes. The system stratifies patients into seven grades, correlating directly with malignancy risk:
-
Grade II: 0% malignancy
-
Grade III: 8.3% malignancy
-
Grade IV: 45% malignancy
-
Grade V: 92% malignancy.
Overall, the system achieved a sensitivity of 96% in detecting malignant lesions, reinforcing its clinical potential.
The researchers included 1,113 patients (mean age = 39 years) in the study, with structured imaging data collected from radiography, computed tomography, and magnetic resonance imaging exams. A nested cross-validation approach ensured robust model training and evaluation. The researchers also compared machine learning predictions against human expert assessments using statistical significance testing.
The study demonstrated that artificial intelligence could match—and potentially enhance—expert-level diagnostic performance in musculoskeletal oncology.
Bone tumors are rare but critical to diagnose accurately, and variability in reporting can lead to unnecessary biopsies or delayed treatment. A standardized, data-driven system like BTI-RADS 2.0 could offer a roadmap for improving patient care across diverse clinical settings.
Author disclosures can be found in the published study.
Source: Radiology
