This study aimed to develop and validate a prediction model that integrates radiomics with clinical characteristics, employing interpretable machine learning methods. The goal was to assist in the differential diagnosis of pheochromocytoma (PHEO) and adrenal adenoma, thereby providing a reference for decision-support information for patients with adrenal tumors. We retrospectively included 107 patients with PHEO and 230 patients with adrenal adenoma, all of whom were pathologically confirmed. Based on contrast-enhanced CT scans, we extracted 1, 316 radiomics features and performed multiple rounds of feature selection to identify those with high discriminative relevance. Then we developed models incorporating these features along with clinical data, utilizing various algorithms including SVM, RF, SGD, KNN, XGBoost, and LightGBM. The diagnostic performance of these models was assessed using receiver operating characteristic (ROC) curves, Decision Curve Analysis, calibration curves, and DeLong tests. Finally, we used the SHapley Additive exPlanations (SHAP) method to interpret the contributions of different features. The clinical-radiomics model demonstrated superior performance, achieving an area under the ROC curve (AUC) of 0. 938. The Decision curve analysis (DCA) indicated that this model was more beneficial than either the clinical models or the radiomics models. Additionally, the SHAP method highlighted the contribution of each feature in the final model, with “log-sigma-1-mm-3DglszmGLNU” identified as the most important feature. Our study showed that the clinical-radiomics model using contrast-enhanced CT could effectively distinguish PHEO from adrenal adenomas.
Li et al. (Thu,) studied this question.