MRI-based Model For Prediction Of The Prognosis In Pituitary Adenomas After Surgery

Part one Nomogram predictive model of extent of resection in pituitary adenoma patients with transsphenoidal surgeryPurpose:To analyze the factors associated with extension of resection in pituitary adenoma patients,and to construct a nomogram to predict whether it could achieve gross total resection.Materials and Methods:In this study,we collected all consecutive patients with pituitary adenomas(confirmed by pathological results)ranged from November 2011 to June 2019 in General Hospital of Eastern Theater Command.All patients underwent transsphenoidal surgery by experienced surgeons and magnetic resonance imaging scans before surgery and 3 months postoperatively.According to MRI images before and after surgery,as well as surgical records from hospital information system,to determine whether there was residual tumor.Therefore,patients are divided into gross total resection(GTR)group and subtotal resection(STR)group.All patients’demographic information and clinical data are documented.By reading pituitary MRI scans before surgery,Knosp grade of pituitary adenomas,maximum tumor diameter(arbitrary plane),maximum horizontal tumor diameter and the intercarotid distance at the intracavernous horizontal C4 segment of the internal carotid artery are also recorded.Predictors for gross total resection of pituitary adenoma were analyzed by using logistic regression model and a nomogram were constructed to predict extent of resection.Patients who underwent transsphenoidal surgery in Taizhou hospital from March 2018 to April 2021 and those in the first affiliated hospital of Xuzhou Medical University were collected as external dataset for validating performance of nomogram model.Results:A total of 384 patients in our hospital are enrolled in our study,including 259 patients in GTR group and 125 in STR group.55 patients in external dataset include 37 patients in GTR group and 18 in STR group.Four risk factors,the maximum tumor diameter,ZPS score,invasion of cavernous sinus and the consistency of tumor,are incorporated in the nomogram model.The AUC value of model in the training dataset is 0.859 and the AUC value in the validation dataset is 0.806.The reliability and clinical practicability of model are confirmed by calibration curves and decision curve analysis.Conclusion:Our nomogram model could well predict extent of resection of pituitary adenoma and aid surgeons to make surgery scheme individually.Part Two A MRI-based radiomics model for predicting postoperative progression/recurrence in pituitary adenomasPurpose:Patients with pituitary adenomas would have a good prognosis after transsphenoidal surgery,while a subset of them may suffer from progression/recurrence(P/R).This study combine clinical features and radiomic features that extracted fom preoperative MR imaging to construct a predictive model for prediction of P/R and risk stratification of postoperative patients.Materials and Methods:We collected all consecutive patients with pituitary adenomas(confirmed by pathological results)ranged from November 2011 to June 2019 in General Hospital of Eastern Theater Command in this study.All patients underwent transsphenoidal surgery by experienced surgeons and magnetic resonance imaging scans before and after surgery.According to MRI images during follow-up,the prognosis of patients are evaluated.In the meantime,patients are divided into P/R group and non-P/R group.The radiomic features are extracted from masks of preoperative MR imaging which were manually delineated layer by layer.Some algorithm are used to selecting importing and robust features,which were combined with clinical features for construction of predictive model,and the performance of the model was evaluated with some indices.Risk factors which were associated with P/R were investigated by univariate and multivariate Cox hazards analysis and Kaplan-Meier graph were plotted.Patients who underwent transsphenoidal surgery in Taizhou hospital from March 2018 to April 2021 and those in the first affiliated hospital of Xuzhou Medical University were collected as external dataset for validating performance of predictive model.Results:A total of 385 patients in our hospital and 44 patients from two other institutions are enrolled in this study,as internal cohort,external cohort respectively.In multiple Cox hazards analysis,subtotal resection was high-risk factor for P/R(p<0.05)with hazard ratios of 26.044.The AUC value of the combined model that was constructed with clinical features and radiomics features was 0.907,0.894,0.826 in training dataset,test dataset and external validation dataset,respectively.Conclusion:Our preliminary combined model showed that radiomic features may have the potential to offer valuable information in predicting P/R in pituitary adenomas after surgery.