Development of a diagnostic model for intradural extramedullary spinal schwannomas and spinal meningiomas based on magnetic resonance imaging radiomics and deep learning

To develop a diagnostic model distinguishing intradural extramedullary (IDEM) spinal schwannomas (SS) and spinal meningiomas (SM) using magnetic resonance imaging (MRI) radiomics and deep learning features. This retrospective study included 68 IDEM-SS and 33 IDEM-SM patients from a tertiary hospital (October 2021–October 2023). T1-weighted enhanced MRI images were analyzed. Radiomics features ( n = 107) were extracted via Pyradiomics; after statistical selection, five key features were retained for model building. Deep learning models (ResNet18, ResNet50, ResNet152) extracted high-dimensional features, further selected via minimum redundancy maximum relevance (mRMR) and least absolute shrinkage and selection operator (LASSO). Both radiomics and deep learning features were classified using four algorithms: logistic regression (LR), support vector machine (SVM), k-nearest neighbors (KNN), and multilayer perceptron (MLP), with tenfold cross-validation and grid search for optimization. Model performance was assessed by the area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and decision curve analysis (DCA). The SVM-based radiomics model achieved an AUC of 0.821. Deep learning models outperformed radiomics, with ResNet50 + MLP achieving the best results (AUC 0.930, accuracy 0.871, sensitivity 0.917, specificity 0.842). DCA indicated robust clinical utility for ResNet-based models. MRI-based deep learning models, especially ResNet50 + MLP, show promising performance in differentiating IDEM-SS from IDEM-SM, potentially aiding clinical decision-making.