MRI-based Radiomics For Preoperative Prediction Of PRL Expression Status In Pituitary Adenomas

ObjectiveTo retrospectively investigate the potential of MRI-based radiomics features in predicting the PRL expression status of pituitary adenomas.MethodsA total of 103 patients who pathologically confirmed pituitary adenomas(02/2013-08/2018)(50 with PRL positive expression and 53 with PRL negative expression)were enrolled segmentation and 1029 radiomics features selection were based on T1-weighted images(T1WI),T2-weighted images(T2WI)and contrast-enhanced T1-weighted images(CE-T1WI),respectively.Features significantly associated with the prolactin expression were selected by using Variance Threshold,Analysis of variance as well as the Least absolute shrinkage and selection operator(Lasso)methods.The 5-fold cross validation strategy and logistic regression model were applied in the machine learning process.The performance of the models was analyzed based on the receiver operating characteristics(ROC)curve and the best model was selected.Combined the radiomics signature and other clinical characters(serological PRL level and age),we built a developed nomogram for the preoperative prediction of PRL expression for pituitary adenomas.ResultsThe radiomics signature containing 19 selected features was significantly associated with PRL expression.In the training group,the radiomics model based on T1WI&T2WI&CE-T1 WI images demonstrated a promising performance(Area Under Curve(AUC): 0.76,95%CI:0.66 to 0.85;specificity: 0.68;sensitivity: 0.72;accuracy: 0.70),which was verified in the validation group(AUC: 0.77,95%CI: 0.65 to 0.89;specificity: 0.66;sensitivity: 0.68;accuracy: 0.67).The prediction results of nomogram besed on radiomics signature,other clinical characters and its combination was(AUC: 0.87,95%CI: 0.80 to 0.93;specificity: 0.81;sensitivity: 0.78;AUC: 0.71,95%CI: 0.61 to 0.81;specificity: 0.89;sensitivity: 0.50;AUC: 0.91,95%CI: 0.86 to 0.97;specificity: 0.87;sensitivity: 0.84).ConclusionRadiomics is a promising approach to the prediction of PRL expression before treatment,providing assistance in the preoperative diagnosis of PRL expression of pituitary adenomas and prognosis for pituitary adenomas.