Abstract： Objective:To explore the effectiveness and feasibility of the machine learning models based on radiomics in the diagnosis of pituitary prolactin macroadenoma.Methods:Totally 122 histologically proven pituitary macroadenoma patients, including 70 cases of pituitary prolactin macroadenoma (PPM) and 52 cases of non-pituitary prolactin macroadenoma (NPPM), were retrospectively recruited. The differences of age, sex, serum prolactin value, bleeding, cystic degeneration and Knosp classification were compared between PPM and NPPM. The pre-processing, delineation of the region of interest and feature extraction of the preoperative axial contrast-enhanced T 1WI image were performed in the 3Dslicer software. The optimal feature set were selected by least absolute shrinkage and selection operator. All patients were randomly divided into the training group ( n=85) and the test group ( n=37) at a ratio of 7∶3. The models were established in the training group by logistic regression and support vector machine (SVM), and then verified by the test group. ROC curves were drawn respectively, and specificity, sensitivity, accuracy and area under the ROC curve (AUC) were calculated. Results:The age [(38±12) years vs . (43±11) years], gender ratio (male/female 50 cases/20 cases vs . 14 cases/38 cases) and prolactin value [366.00 (117.75, 1 156.25)μg/L vs . 47.25 (32.68, 62.40) μg/L] of patients with PPM and NPPM were statistically different ( P<0.05). The AUC values of logistic regression and SVM in the training group were 0.936 and 0.946, and the AUC values of the test group were 0.768 and 0.774, respectively. The diagnostic accuracy of logistic regression and SVM in the training group were 88.2% and 91.8%, and the accuracy of the test group were 73.0% and 77.8%. Conclusion:The machine learning models based on the radiomics can predict the pituitary prolactin macroadenoma well with a high accuracy.