Abstract
OBJECTIVE: To investigate the clinical value of preoperative prediction of pituitary neuroendocrine tumor（PitNET） consistency based on multiparametric magnetic resonance imaging (mpMRI) radiomics. PATIENTS AND METHODS: The clinical data of 137 patients with PitNET who underwent preoperative mpMRI were retrospectively analyzed and classified into soft and hard according to the consistency of the PitNET tumor with the surgical records of neurosurgeons. The patients were randomly divided into two sets: a training set (n = 108) and an internal validation set (n = 29). Single and multifactorial factors were used to analyze clinical high-risk risk factors and establish clinical models. Using the logistic regression (LR) classifier to construct radiomics signature based on 2D and 3D region of interest(ROI), respectively. Combined with clinical characteristics and radiomics features, a combined clinical-radiomics model was constructed, and the nomogram was drawn. The robustness and accuracy of the prediction model were tested by using multi-center clinical data as an external validation set. RESULTS: 4224 and 5061 radiomics features were extracted based on 2D ROI and 3D ROI, and 28 and 15 predictive features were selected. Among the radiomics signature, the 3D-multi (T1WI + T2WI + CE-T1) radiomics signature constructed based on 3D ROI has high prediction efficiency. The area under curve(AUC) values in the training set and the internal validation set are 0.793 (95% confidence interval (CI): 0.711-0.859) and 0.798 (95% CI: 0.643-0.942), respectively. Among the combined clinical-radiomics models, the 2D&3D ROI model have the highest prediction efficiency, with the AUC values of 0.894 (95% CI: 0.832-0.942) and 0.813 (95% CI: 0.667-0.926) in the training set and the internal validation set, respectively. CONCLUSION: In this study, the mpMRI (T1WI+T2WI+CE-T1) radiomics model could effectively and accurately predict the consistency of pituitary neuroendocrine tumor before Surgery, and the prediction efficiency of the radiomics model based on 2D and 3D ROI is different.