Differentiating Schmorl's nodes from osteolytic bone metastases: diagnostic performance of conventional CT features and CT-based models incorporating radiomics and CT features.

Abstract

To evaluate conventional CT features for differentiating Schmorl's nodes from osteolytic bone metastases and to compare their diagnostic performance with that of CT-based models incorporating radiomics and CT features. In this retrospective study, 79 Schmorl's nodes and 71 osteolytic bone metastases-well-defined, geographic, and purely osteolytic lesions located at the vertebral endplates on abdominal and chest CT images-were randomly divided into two groups: a training set (n = 100) and a test set (n = 50). Subjective image analysis was performed to identify significant discriminative CT features and to construct a CT feature-based model. Random forest models were constructed for the radiomics, CT feature-based, and combined radiomics-CT feature approaches, followed by tenfold cross-validation. Diagnostic performance was evaluated using the area under the receiver operating characteristic curve (AUC), and AUCs were compared. Multivariate analysis identified distribution (odds ratio [OR], 11.109; 95% confidence interval [CI], 2.417-51.065; p < 0.01) and thickness (OR, 145.431; 95% CI, 14.761-1432.896; p < 0.001) of the sclerotic margin as significant independent factors for differentiation. The CT feature-based model, combined model, and combination of significant CT features demonstrated high diagnostic performance, with AUCs of 0.912, 0.912, and 0.901, respectively (all p < 0.001), without significant differences among them. CT feature-based and combined models were effective in differentiating Schmorl's nodes from osteolytic bone metastases on CT imaging. The combination of significant CT features-specifically, a completely sclerotic margin and a margin thickness equal to or greater than the adjacent cortex-showed comparable performance, offering a simple diagnostic and practical alternative.