Abstract
OBJECTIVE: To develop and externally validate a multiparametric magnetic resonance imaging (MRI)-based radiomics nomogram for preoperative differentiation of primary from metastatic lumbar spine tumors. METHODOLOGY: 200 patients were divided into training (n=100) and independent external validation (n=100) cohorts. Radiomics features from T1WI, T2WI, FS-T2WI were filtered and reduced via LASSO to construct Radscore; a combined nomogram integrating Radscore and clinical variables was evaluated. RESULTS: The combined nomogram demonstrated excellent discriminatory ability in the independent external validation cohort, with an area under the curve (AUC) of 0.921 (95% confidence interval [CI]: 0.838-0.970). Its performance was significantly superior to that of the clinical variables-only model (AUC: 0.732, P < 0.001) and the Radscore-only model (AUC: 0.880, P = 0.028), achieving a sensitivity of 85% and a specificity of 87%. Univariate and multivariate logistic regression analyses identified the Radscore, age > 60 years, and serum alkaline phosphatase (ALP) > 120 U/L as independent predictors for differentiating primary from metastatic lumbar spine tumors. The nomogram exhibited good calibration (Hosmer-Lemeshow test, P = 0.62). Decision curve analysis (DCA) confirmed its clinical utility by showing a higher net benefit across a wide range of threshold probabilities compared to default strategies. CONCLUSION: A radiomics nomogram integrating multiparametric MRI features and key clinical factors was successfully developed and externally validated. It serves as an effective, non-invasive auxiliary tool for preoperative differentiation of primary from metastatic lumbar spine tumors, with potential for clinical translation.