Akademik Veri Yönetim Sistemi
Genel Tıp Dergisi, cilt.36, sa.1, ss.1-7, 2026 (Scopus, TRDizin)
Aim: Osteosarcoma (OS) and chondrosarcoma (CS) are common malignant bone tumors with overlapping clinical and radiological features, making accurate differentiation challenging. Biopsy, the gold standard, is invasive and prone to sampling errors. This pilot study aimed to evaluate whether radiomic features from T2-Weighted magnetic resonance imaging (MRI) can non-invasively distinguish OS from CS.
Materials and Methods: This retrospective study included 29 histopathologically confirmed patients (OS=15, CS=14). Pre-treatment T2WI scans were acquired on 1.5T/3T MRI scanners. An experienced radiologist manually segmented tumors, excluding hemorrhage, necrosis, or edema. 107 radiomic features extracted. Features were ranked by combining statistical testing with Principal Component Analysis (PCA) loadings, and the top five were retained. A baseline linear discriminant analysis (LDA) was evaluated with five-fold stratified cross-validation, and an advanced pipeline (standardization, SMOTE, kernel PCA, LDA, and RBF-SVM) was optimized by grid search and cross-validation on the training set, and subsequently tested on an independent split. Model performance was evaluated with accuracy, F1 score, balanced accuracy, ROC–AUC, and confusion matrices, expressed as misclassification between tumor subtypes.
Results: Of 107 extracted radiomic features, shape and texture metrics were most highly prioritized. A baseline LDA achieved modest cross-validated accuracy (0.53) with respect to OS but showed significant class separation. The optimized pipeline improved test performance, reaching a balanced accuracy of 0.83, F1 score of 0.86, ROC–AUC of 0.78, and 100% sensitivity with respect to OS at the optimal threshold.
Conclusions: Radiomic analysis of pre-treatment T2WIs shows promise for differentiating osteosarcoma from chondrosarcoma, achieving high sensitivity and balanced accuracy, and may serve as a complementary tool to conventional diagnosis in musculoskeletal oncology.