Prediction Of Pituitary Adenomas Aggressiveness And Ki-67 Index Based On MR Images And Machine Learning

Objective:In this study,quantitative radiomics characteristics were extracted from preoperative pituitary adenomas areas and peritumoral areas,and a radiomics model based on machine learning technology was built to explore the feasibility of predicting the invasiveness and Ki-67 index of pituitary adenomas.Methods:A total of 198 patients with pituitary adenomas confirmed by pathological examination from July 2018 to July 2021 in Shandong Provincial Hospital were retrospectively collected.The preoperative MR images and postoperative immunohistochemical results of the above patients were obtained-According to intraoperative findings,there were 109 invasive pituitary adenomas and 89 non-invasive pituitary adenomas.According to the results of immunohistochemical examination of surgical specimens,there were 97 cases with a high Ki-67 labeling-index(>3%)and 101 cases with low Ki-67 labeling index(≤3%).For the two classification tasks of invasiveness and Ki-67 index of pituitary adenomas,all the enrolled cases were divided into training set and validation set with a ratio of 9:1,which were used for training and evaluating machine learning models.Using the coronal CE-T1 sequence as a reference,the tumor region of interest(ROI)was delineated on the 3D Slicer software,and the peritumoral ROI region was obtained through further processing.Feature extraction was performed on all the above ROIs using PyRadiomics software,including first-order histogram features,shape features and texture features.Variance threshold and Least Absolute Shrinkage and Selection Operator(LASSO)are applied to select radiomics characteristics that are highly relevant to the two classification tasks gradually.We used the above radiomics features to build prediction models based on support vector machines(SVMs).The model was validated using the validation set,the receiver operating characteristic curve(ROC)was drawn,and the predictive performance of the two models was evaluated using the area under the curve(AUC),accuracy,sensitivity,precision and F1-Score as evaluation metrics.Results:(1)A total of 19 highly correlated radiomics characteristics were selected in the model predicting the aggressiveness of pituitary adenomas with an area under the curve(AUC)of 0.73(95%CI:0.75-1.00)in the validation cohort,The accuracy was 0.85,the precision was 0.93,the recall was 0.87 and the F1-Score was 0.90.(2)A total of 36 highly correlated radiomics characteristics were selected in the model for predicting the Ki-67 index of pituitary tumors,and the model’s area under the curve(AUC)in the validation cohort was 0.76(95%CI:0.57-0.94),the precision was 0.75,the precision was 0.80,the recall was 0.73 and the F1-Score was 0.76.Conclusion:The machine learning model of tumor and peritumoral area based on MR images can predict the invasiveness and Ki-67 expression index of pituitary adenoma,which can provide more valuable reference for the formulation of pituitary adenoma surgery plan and prognosis prediction.