Objective: The aim of this study was to develop a clinical-radiomics fusion model to predict the efficacy of surufatinib on hepatic metastatic neuroendocrine neoplasms (HM-NENs) through chest-unenhanced computed tomography (CUE CT).
Methods: This study included 76 HM-NEN patients (131 hepatic metastases) treated with surufatinib. SlicerRadiomics was employed to extract radiomics features of hepatic metastases from CUE CT scans. The Least Absolute Shrinkage and Selection Operator (LASSO) was used to select radiomics features and calculate a Radiomics score (Radscore). Multivariable logistic regression analysis was utilized to create a clinical-radiomics fusion model, which included clinical characteristics and Radscore, displayed as a nomogram. The area under the receiver operating characteristic curve (ROC) was used to assess model performance, and internal validation was performed using the bootstrap resampling approach.
Results: After multivariate logistic regression analysis, Radscore, the diameter of the hepatic metastasis, number of hepatic metastases, and extrahepatic metastasis were independent prognostic factors (p＜0.05). The area under the curve (AUC) of the fusion model was 0.926（95%CI:0.881-0.971）. The AUC verified by bootstrap was 0.928（95%CI:0.881-0.965）, indicating a good performance of the fusion model.
Conclusion: The clinical-radiomics model developed based on CUE CT can effectively identify HM-NEN patients who are sensitive to surufatinib. This model not only avoids contrast agent exposure required for enhanced scanning, thereby reducing the risk of hepatorenal function injury and potentially toxic side effects but also reduces the economic cost of imaging examinations for patients. Additionally, by accurately predicting treatment efficacy, this model is expected to provide a noninvasive and efficient new tool for clinical individualized treatment decisions.