Radiomics-based identification of benign and malignant orbital lesions using contrast-enhanced CT imaging

This study aimed to evaluate the value of contrast-enhanced computed tomography (CT) imaging radiomics in distinguishing malignant lesions from benign ones in the orbit. A retrospective analysis was conducted on CT imaging data from 139 patients with orbital tumor lesions, all of whom underwent contrast-enhanced CT scans within 2 weeks before diagnosis. Of these, 45 cases were benign lesions and 94 were malignant lesions. Radiomic features were extracted from the contrast-enhanced CT images, and 12 features were selected through the minimum redundancy maximum relevance and least absolute shrinkage and selection operator regression methods. The selected features were used to build models using logistic regression, Naive Bayes Classifier (NaiveBayes), support vector machine (SVM), Extra Trees Classifier (ExtraTrees), and multilayer perceptron, with the best-performing model identified. Multivariate logistic regression was employed to identify clinical risk factors for malignant orbital lesions, and a nomogram model was developed by combining radiomic features and clinical variables. The predictive performance of each model was evaluated using the area under the receiver operating characteristic curve. Among the 3 machine learning models, the SVM model demonstrated the best predictive performance and robustness across datasets. Therefore, the SVM model was used to construct the nomogram. The nomogram achieved area under the receiver operating characteristic curve values of 0.957 and 0.833 in the training and testing cohorts, respectively, both of which were higher than 0.80. The performance of the nomogram was significantly superior to that of the clinical model (De-long test, P < .05), but no statistically significant difference was observed when compared to the radiomics model (De-long test, P > .05). Contrast-enhanced CT radiomics can effectively differentiate between malignant and benign orbital lesions. Both the nomogram and radiomics models exhibited high predictive performance, offering valuable insights for clinical decision-making.