CT Radiomics-based Machine-learning Classification of Atypical Cartilaginous Tumors and Appendicular Chondrosarcomas

Presentation Format: Oral presentation.

Purpose or Learning Objective: To investigate the diagnostic performance of computed tomography (CT) radiomics-based machine learning for the classification of atypical cartilaginous tumors and higher grade chondrosarcomas of long bones.

Methods or Background: A total of 120 patients with surgically treated and histology-proven cartilaginous bone tumors were retrospectively included at two tertiary tumor centers. The training cohort consisted of 84 CT scans from center 1 (n = 55 G1 or atypical cartilaginous tumors; n = 29 G2–G4 chondrosarcomas). The external test cohort consisted of the CT component of 36 positron emission tomography-CT scans from center 2 (n = 16 G1 or atypical cartilaginous tumors; n = 20 G2–G4 chondrosarcomas). All tumors were manually segmented on preoperative CT, using the axial image showing the maximum lesion extension. First-order, shape-based, and matrix features were extracted. After dimensionality reduction, a machine-learning classifier (LogitBoost) was tuned on the training cohort using 10-fold cross validation and tested on the external cohort. In patients from center 2, the classifier's performance was compared with the preoperative biopsy using the McNemar test.

Results or Findings: The classifier had 81% (area under the curve [AUC] = 0.89) and 75% (AUC = 0.78) accuracy in identifying the cartilaginous tumors in the training and external test cohorts, respectively. Its accuracy in classifying atypical cartilaginous tumors and higher grade chondrosarcomas was 84% and 78% in the training cohort, and 81% and 70% in the external test cohort, respectively. Preoperative biopsy had 64% (AUC = 0.66) accuracy in center 2 (p = 0.289).

Conclusion: Machine learning showed good accuracy in classifying atypical cartilaginous tumors and higher grade chondrosarcomas of long bones based on CT radiomic features and could prove a valuable aid in preoperative assessment.

Publikationsverlauf

Artikel online veröffentlicht:
03. Juni 2021