The Value Of MR-radiomics In Identifying Pathological Type Of Pituitary Adenomas

ObjectivePituitary adenomas is the most common tumor type of pituitary gland and saddle area and prolactinomas is the most common pituitary adenomas;the accurate identification of prolactinomas and other types of pituitary adenomas is vital to choice of therapy.This study was,using radiomics,to develop and validate a magnetic resonance imaging-based radiomic model、a clinic model and a clinic-radiomic model to investigate its value in predicting pathological type of pituitary adenomas.Material and MethodsOne hundred and forty-two patients with pituitary adenomas were retrospective enrolled and divided into lactotroph adenomas group(n=79)and non-lactotroph adenomas group(n=63)according to immunohistochemical staining.All patients were randomly divided into training cohorts(n=99)and testing cohorts(n=43)with 55 lactotroph adenomas patients in training cohorts and 24 in validation cohorts.Radiomic features were extracted from their MR images and radiomic models(model 1 using T1WI+T2WI;model 2 using T1WI+T2WI+CET1WI)were built using radiomics.Subsequently,multivariable logistic regression analysis was used to select the independent risk factors and a clinic model was built using support vector machine.A clinic-radiomic model incorporating the selected radiomic features and independent risk factors was constructed using support vector machine.ResultThe radiomic model 1 based on 10 radiomic features showed an AUC、sensitivity、specificity and accuracy of 0.753、78.2%、59.1%、69.6%in the training cohorts and 0.757、79.2%、57.9%、69.7%in testing cohorts.The radiomic model 2 based on 16 radiomic features,showed an AUC、sensitivity、specificity and accuracy of 0.834、89.1%、61.4%、76.8%in the training cohorts and 0.800、79.2%、52.6%、67.4%in testing cohorts.The clinic model based on age、hyperprolactin、cavernous sinus involvement、shape showed an AUC、sensitivity、specificity and accuracy of 0.829、67.3%、77.3%、71.7%in the training cohorts and 0.757、62.5%、68.4%、65.1%in testing cohorts.The clinic-radiomic model incorporating radiomic model and clinic model showed an AUC、sensitivity、specificity and accuracy of 0.943、92.7%、86.4%、89.9%in the training cohorts and 0.820、83.3%、57.9%、72.1%in testing cohorts.ConciusionThe clinic-radiomic model through incorporating age、hyperprolactin、cavernous sinus involvement、shape and radiomic features showed good predictive ability,which may have important value in determining individual treatment strategies.