Abstract / Summary
Abstract Objective To investigate the diagnostic value of machine learning models based on CT radiomics in differentiating adrenal metastases (AMs) from lipid-poor adenomas (LPAs). Methods A total of 212 patients with AMs and 213 patients with LPAs were retrospectively analyzed based on CT images from the First Affiliated Hospital of Bengbu Medical University (Center 1) and Tongde Hospital of Zhejiang Province (Center 2). Specifically, images from Center 1 (178 AMs and 189 LPAs) constituted the training and internal validation cohorts, while images from Center 2 (34 AMs and 24 LPAs) served as the external validation cohort. Radiomic features were extracted from unenhanced CT (CTU), arterial phase (CTA), and venous phase (CTV) images. Four machine learning models, including Adaptive Boosting (ADA), Decision Tree (DT), K-Nearest Neighbors (KNN), and Logistic Regression (LR), were constructed after dimensionality reduction and feature selection. To compare diagnostic performance, we constructed a clinical baseline model based on routine radiological and clinical indicators, as well as a multiphasic feature fusion model. Diagnostic performance across different models was compared via receiver operating characteristic (ROC) curves and the DeLong tests. Results Among the four models, the LR model based on CTV images (LR-CTV) demonstrated the best diagnostic performance. In the internal validation set, the Area Under the Curve (AUC), sensitivity, specificity, and accuracy were 0.96, 93.3%, 92.2%, and 92.8%, respectively. In the external validation set, the corresponding metrics were 0.96, 86.6%, 90.6%, and 87.9%. LR models significantly outperformed conventional clinical baseline models across all three phases and in both validation cohorts (all p < 0.05). No statistically significant difference was found between the LR-fusion model and its optimal single-phase counterpart (LR-CTV) in either cohort (all p > 0.05). Conclusion The LR-CTV model exhibited excellent efficacy and good clinical applicability in differentiating AMs from LPAs.