Radiomics and Radiogenomics for Differentiating Skull Base Chordoma and Chondrosarcoma: A Systematic Review and Diagnostic Meta-Analysis

Background: Skull base chordomas and chondrosarcomas are rare primary bone tumors with overlapping radiographic features but distinct clinical management pathways. Accurate preoperative differentiation is essential, as chordomas often require more aggressive surgical strategies and carry higher recurrence risks. Radiomics, with or without radiogenomic integration, offers a quantitative imaging approach that may enhance diagnostic precision. We conducted a systematic review and proportional meta-analysis to evaluate the diagnostic performance of radiomics-based methods in differentiating skull base chordomas from chondrosarcomas. Methods: PubMed and Embase were searched through 2025 using the query: (chordoma OR chondrosarcoma) AND (skull base) AND (radiomics OR machine learning OR deep learning) AND (diagnosis OR differentiation). Studies were included if they applied radiomics or radiogenomic models to imaging for differentiation of skull base chordomas versus chondrosarcomas. Prognostic, therapeutic, histopathologic, and review studies were excluded. Data extraction focused on sample size, tumor distribution, imaging modality, classifier type, and diagnostic metrics. A proportional meta-analysis was performed in R version 4.4.2 under a random-effects model, expressed as events/total with 95% confidence intervals (CIs). AUCs were synthesized narratively due to incomplete variance reporting. Results: Three studies with a combined total of 288 patients were included in the analysis, comprising 214 chordomas (74%) and 74 chondrosarcomas (26%). All studies employed multiparametric MRI with machine learning classifiers. The pooled diagnostic accuracy of radiomics-based models was 0.80 (95% CI: 0.70–0.90), although there was moderate heterogeneity between studies (I 2 =77%). Sensitivity for detecting chordomas was high, with a pooled estimate of 0.90 (95% CI: 0.86–0.95) across 184 chordoma cases, and no heterogeneity observed (I 2 =0%). In contrast, the pooled specificity for identifying chondrosarcomas was lower, at 0.68 (95% CI: 0.57–0.79) across 65 cases, again with no heterogeneity (I 2 =0%). Analysis of the case distribution confirmed that chordomas were more frequently represented across cohorts, with a pooled proportion of 0.74 (95% CI: 0.69–0.79), while chondrosarcomas accounted for a pooled proportion of 0.26 (95% CI: 0.21–0.31). Narrative synthesis of AUC values demonstrated moderate to excellent discriminative performance, ranging from 0.78 to 0.93 across studies, with Li et al. reporting an AUC of 0.87, Yamazawa et al. reporting 0.93, and Stastna et al. reporting 0.78. Conclusion: Radiomics-based diagnostic models achieved strong pooled accuracy and high sensitivity for detecting chordomas, though specificity for chondrosarcomas was more modest. The underlying sample distribution highlighted the predominance of chordoma cases in published studies, which may influence model performance and generalizability. Reported AUCs support good overall discriminative ability, but methodological heterogeneity, small retrospective cohorts, and lack of external validation remain important limitations. Future efforts should prioritize multicenter collaborations, standardized radiomic pipelines, and radiogenomic integration to improve non-invasive differentiation of skull base chordomas and chondrosarcomas. Publication History Article published online: 27 February 2026 © 2026. Thieme. All rights reserved. Georg Thieme Verlag KG Oswald-Hesse-Straße 50, 70469 Stuttgart, Germany