The Role of Artificial Intelligence Algorithm in Predicting the Prognosis in Prolactinomas

Abstract

Objective To test the utility of the artificial learning algorithms using magnetic resonance (MR) images of the pituitary gland in predicting the prognosis of prolactinoma. Methods This single-center, retrospective study was conducted in the Pituitary Center of a tertiary care university hospital. A total of 224 images derived from 38 patients with treatment-refractoryprolactinoma, 23 patients with prolactinoma remission and 51 healthy individualswere used. Pituitary MRI protocols are of three sequences: T1-weighted imaging (T1WI), contrast-enhanced T1WI (CE-T1), and T2-weighted imaging (T2WI). A machine learning algorithm that includes image filtering and classification. Data were classified with support vector machine. Results No difference was found between the refractory and the remission groups in terms of age, sex, education, the baseline prolactin level and radiological features. Images were classified with a support vector machine; area under curve (AUC), accuracy, sensitivity and specificity of 0.90 (95% confidence interval, 0.679-1), 91.6%, 91.7%, 88.3%, respectively. Conclusion These results indicate that a new image of unknown nature can be correctly identified with the specified percentages.