Prediction of the Ki-67 expression level and prognosis of gastrointestinal stromal tumors based on CT radiomics nomogram

Purpose To build and validate a radiomics nomogram integrated with the radiomics signature and subjective CT characteristics to predict the Ki-67 expression level of gastrointestinal stromal tumors (GISTs). Moreover, the purpose was to compare the performance of pathological Ki-67 expression level with predicted Ki-67 expression level in estimating the prognosis of GISTs patients. Methods According to pathological results, patients were classified into high-Ki-67 labeling index group (Ki-67 LI ≥ 5%) and low-Ki-67 LI group (Ki-67 LI < 5%). Radiomics features extracted from contrast-enhanced CT(CECT) images were selected and classified to build a radiomics signature. A combined model was built by incorporating radiomics signature and determinant subjective CT characteristics using multivariate logistic regression analysis. The diagnostic performance of the radiomics signature, subjective CT model and combined model were explored by receiver operating characteristic (ROC) curve analysis and Delong test. The model with best diagnostic performance was then set up for the prediction nomogram. Recurrence-free survival (RFS) rates were compared utilizing Kaplan–Meier curve. Results The generated combined model yielded the best diagnostic performance with area under the curve (AUC) values of 0.738 [95% confidence interval (CI): 0.669–0.807] and 0.772 (95% CI 0.683–0.860) in the training set and testing set respectively. The nomogram based on the combined model demonstrated good calibration in the training set and testing set (both P > 0.05). Patients of high-Ki-67 LI group predicted by our nomogram had a poorer RFS than patients of low–Ki-67 LI group ( P < 0.0001). Conclusion This radiomics nomogram based on CECT had a satisfactory performance in predicting both the Ki-67 expression level and prognosis noninvasively in patients with GISTs, which may serve as an effective imaging tool that can assist in guiding personalized clinical treatment.