Multiparameter MRI-based clinical-radiomic model to differentiate prolactinoma from hyperprolactinemic nonfunctioning...

It is crucial to differentiate between prolactinoma and nonfunctioning pituitary adenoma (NFPA) prior to treatment because their preferred therapeutic approaches differ. The optimal treatment for NFPA causing compressive symptoms is transsphenoidal surgical debulking, whereas most prolactinomas respond favorably to dopamine agonist therapy. Both can manifest as hyperprolactinemia; however, existing guidelines solely rely on measurement of the serum prolactin level, which has been proven insufficient. The aim of this study was to distinguish prolactinoma and NFPA with moderate hyperprolactinemia using a clinical-radiomic model before treatment.

A total of 203 patients (174 female, median age 33.0 years) with moderate hyperprolactinemia were diagnosed with prolactinoma (n = 113) or NFPA (n = 90) and divided into training and validation sets retrospectively. An elastic net algorithm was used to screen radiomic features, leading to the establishment of a fusion radiomic model. Subsequently, a clinical-radiomic model was developed by integrating the radiomic model with the significant clinical features for individualized predictions. The calibration, discrimination, and clinical applicability of the models were evaluated. Additionally, the performance of the final selection model was validated using a dataset of 32 patients from other centers.

The fusion radiomic model was developed using 3 significant radiomic features, achieving areas under the curve (AUCs) of 0.900 in the training set and 0.890 in the validation set. The clinical-radiomic model was created by integrating the fusion radiomic model with 3 additional clinical features (age, sex, and serum prolactin level). This model demonstrated excellent recognition and calibration capabilities, with AUCs of 0.94, 0.93, and 0.90 in the training, validation, and external multicenter validation sets, respectively. The decision curve analysis indicated that the constructed models had substantial clinical applicability for patients with prolactinoma and NFPA.

Our clinical-radiomic model exhibited high sensitivity and exceptional performance in the differential diagnosis of prolactinoma and NFPA with hyperprolactinemia. This model offers significant potential for the noninvasive development of personalized diagnostic and treatment strategies for individuals with these conditions.