Enhanced Computed Tomography/Magnetic Resonance Imaging Focal Bone Tumor Classification with Machine Learning–based Stratification: A Multicenter Retrospective Study

Purpose or Learning Objective: (1) To evaluate a machine learning–based approach for differentiating between benign and malignant focal bone lesions, and (2) to propose a Bone Tumor Imaging Reporting and Data System 2.0 for further risk stratification.

Methods or Background: This retrospective multicenter trial (NCT04884048) included patients with solitary bone tumors undergoing computed tomography/radiography/magnetic resonance imaging at 10 centers from November 2009 to March 2022. Patients were divided into a training and testing dataset. Predefined radioclinical features were extracted. The training dataset was considered for bootstrapped chi-square feature selection, and extreme gradient boosting (XGBoost) classifiers were optimized using nested cross-validation. Continuous classifier outputs were thresholded to stratify patients into seven malignancy risk classes (Bone Tumor Imaging Reporting and Data System 2.0), and malignancy rates were assessed for the test set. XGBoost and human expert performances were compared using the Wilcoxon signed rank test with a significance level of 0.05.

Results or Findings: In total, 1,113 patients (623 men, mean age 39 years ± 22 [standard deviation]) were included, 298 in the training and 815 in the test dataset. A total of 27 of 80 (34%) multimodal features were selected based on analysis. Best classification performances were achieved by an XGBoost model trained on 27 features, with an F1 score of 0.81 (95% confidence interval [CI] 78–84). This model performed slightly inferior to 28 experienced radiologists, who demonstrated an F1 score of 0.83 (95% CI 0.80–0.85; P < 0.001). The Bone Tumor Imaging Reporting and Data System 2.0 risk grades 2 to 5 were associated with malignancy rates of 0% (95% CI 0–0; 0/102), 8% (95% CI 4–13; 14/168); 45% (95% CI 39–50; 121/271), and 92% (95% CI 89–95; 252/274), respectively, identifying malignant lesions with a sensitivity of 96% (95% CI 94–98; 373/387).

Conclusion: Machine learning algorithm and risk stratification system achieved accurate and standardized bone tumor malignancy grading.

Publikationsverlauf

Artikel online veröffentlicht:
02. Juni 2025