Radiomics Analysis Based On Preoperative MRI Allows For Precise Prediction Of Immunohistochemical Classification In Pituitary Adenomas

Objective:To investigate the value of radiomics in predicting immunohistochemically classify pituitary adenoma subtypes by MRI before surgery.Methods:One hundred and thirty-six patients who immunohistochemically confirmed pituitary adenomas at the hospital from February 2021 to August2022 were retrospectively enrolled,including thirty-five patients with T-box pituitary transcription factor(TPIT)family tumors,fifty-nine patients with pituitary transcription factor 1(PIT-1)family tumors,and forty-two patients with steroidogenic factor 1(SF-1)family tumors.The data set was divided into the test set in a ratio of 8: 2 with 20% of the data set to obtain a test set of 28cases(PIT-1: TPIT: SF-1=12: 7: 9)and a train set of others.The regions of interest(ROIs)were delineated by each layer of pituitary adenoma lesions in contrast enhancement T1-weighted imaging(CE-T1WI)and T2-weighted imaging(T2WI)sequences using 3D-Slicer software.Radiomics feature extraction is accomplished based on the ROIs that had been sketched.A total of2632 radiomics features were obtained.The features with the highest weight from CE-T1 WI and T2 WI sequences were filtered using analysis of variance followed by least absolute shrinkage and selection operator(LASSO),respectively.Seven radiomics features from CE-T1 WI and five from T2 WI are obtained.Finally,three imaging models(decision tree,random forest and extreme gradient boosting)were structured and AUC and accuracy were used to evaluate the performance of these models.Results:In the test sets of two MRI sequences,the random forest model showed high performance(accuracy0.750 in CE-T1 WI,AUC0.957,accuracy0.786 in T2 WI,AUC0.943),while the decision tree(accuracy0.536 in CE-T1 WI,AUC0.758,accuracy0.464 in T2 WI,AUC0.695)showed poor performance.In the CE-T1 WI sequence,the accuracy of the extreme gradient boosting model is0.786 in the test set,0.870 in the training set,0.679 and 0.843 in the T2 WI sequence,resulting in serious overfitting.Conclusions:In this study,a robust radiomics model was constructed to precisely predict immunohistochemically classify pituitary adenoma subtypes.The random forest model with radiomics features based on CE-T1 WI and T2 WI had better performance than the decision tree model and the extreme gradient boosting model,and the model had a good performance in the task of predicting immunohistochemical subtypes of pituitary adenomas and proved that radiomics analysis could supply useful prediction technique.