The Value Of MRI Radiomics In The Diagnosis Of Pituitary Microadenoma And Its Subtypes

PART 1 The value of MRI radiomics in the diagnosis of pituitary microadenomaObjective:To explore the application value of MRI radiomics in the diagnosis of pituitary microadenoma.Methods:The complete data of 95 patients with pituitary microadenoma and 60 normal pituitary glands diagnosed by clinical follow-up,pathology and diagnostic treatment from January 2015 to December 2020 in the First Affiliated Hospital of Wannan Medical College were retrospectively collected.The subjects were randomly divided into training group（109cases）and verification group（46 cases）at a ratio of 7:3.All patients underwent pituitary routine scan and dynamic enhanced scan.The ITK-SNAP software was used to manually outline the ROI layer by layer along the tumor edge on the T1WI,T2WI and CE＿T1WI images,then three-dimensional fusion of these ROIs.The minimum redundancy maximum redundancy（m RMR）and the least absolute shrinkage and selection operator（LASSO）methods were used to reduce the dimensionality of texture features and establish radiomics signature.In clinical and MRI characteristics,continuous variables were compared using two independent sample t-test or Mann-Whitney U test and categorical variables were compared using chi-square test or fisher’s exact test.Multi-factor logistic regression were used to establish personalized model that included radscore and MRI feature.The Hosmer-Lemeshow was used to test the fitting effect of the model,and draws receiver operating curve ROC to evaluate MRI feature models,radiomics models and personalized models in diagnosis efficacy in pituitary microadenoma.Delong test was used to compare the ROC curve AUC of each model.And evaluate its clinical application valuethrough thedecision curve.Results:1.Slope_max,TTP,T1WI signal intensity ratio,T2WI signal intensity ratio and T1WI enhanced signal intensity ratio were significantly different between the training group and the verification group（P<0.05）.ER_maxwas statistically significant difference in the training group（P=0.03）,and there was no statistically significant difference in the validation group（P=0.08）.There was no statistically significant difference in age between the training group and the validation group（P values were 0.99,0.92）.The AUC（95%CI）of the ROC for the MRI feature model constructed by combining ER_max,Slope_max,TTP,T1WI signal intensity ratio,T2WI signal intensity ratio,and T1WI enhanced signal intensity ratio in the training group and theverificationgroup,were0.95（95%CI:0.90～1.00）and0.93（95%CI:0.85～1.00）.2.After de-redundancy by m RMR,and dimensionality reduction by LASSO,14,13,15,15texture features were selected from T1WI,T2WI,enhanced T1WI and joint sequence.According to these characteristics,the radscore of each patient is established by the linear combination of the weighting coefficient products corresponding to these characteristics.According to ROC analysis,the combined radiomics model is the most effective in diagnosing pituitary microadenoma,the AUC of the ROC in training and validation group was0.90（95%CI:0.83～0.96）and 0.79（95%CI:0.62～0.95）.The AUC（95%CI）of the ROC for T1WI radiomics model in training and validation group was 0.82（95%CI:0.72～0.91）and 0.77（0.60～0.94）.The AUC（95%CI）of the ROC for T2WI radiomics model in training and validation group was 0.79（95%CI:0.69～0.91）and 0.78（95%CI:0.59～0.97）.The AUC（95%CI）of the ROC for T1WI enhanced radiomics model in training and validation group was 0.81（95%CI:0.71～0.91）and0.78（95%CI:0.60～0.96）.3.Combining MRI features and radscore to constructed personalized model.After ROC analysis,the AUC of the model in the training group and the validation group were 0.98（95%CI:0.96～1.00）and 0.99（95%CI:0.98～1.00）.ER_max,Slope_max,TTP,T1WI signal intensity ratio,T2WI signal intensity ratio,T1WI enhanced signal intensity ratio,and combined radscore are independent risk factors for the diagnosis of pituitary microadenoma.By Hosmer-Lemeshow test,the model has good goodness of fit in the training group and the validation group（p>0.05）.The AUC difference between the personalized model and the MRI feature model,the combined radiomics model was statistically significant by Delong test（P<0.05）.DCA demonstratedthatthat ithasgoodclinicalbenefits.Conclusion:1.ER_max,Slope_max,TTP,T1WI signal intensity ratio,T2WI signal intensity ratio and T1WI enhanced signal intensity ratio are of significance for the diagnosis of pituitary microadenoma,and MRIfeaturemodel hashighdiagnosticvalue.2.Among the radiomics models,the combined radiomics model is superior to the three single sequence radiomics models,which has high diagnostic value,and is not as efficient as the MRI featuremodel.3.The combination of MRI features and combined radscore to build personalized model is better than the combined radiomics model and the MRI feature model,and both have better clinicalbenefits.PART 2 The value of MRI radiomcs in the differentiation of microprolactinomas and non-microprolactinomasObjective: To explore the value of MRI imaging in the differentiation of microprolactinomas andnon-microprolactinomasMethods:A retrospective analysis of 95 patients diagnosed with pituitary microadenoma through clinical follow-up,pathology anddiagnostic treatment in our hospital from January 2015 to December 2020,including 62 microprolactinomas and non-microprolactinomas 33 cases;the patients were randomly divided into training group（68 cases）and verification group（27 cases）according to the ratio of 7:3.All patients underwent pituitary routine scan and dynamic enhanced scan.The ITK-SNAP software was used to manually outline the ROI layer by layer along the tumor edge on the T1 WI,T2WI and CE_T1WI images,then three-dimensional fusion of these ROIs.The minimum redundancy maximum redundancy（m RMR）and the least absolute shrinkage and selection operator（LASSO）methods were used to reduce the dimensionality of texture features and establish radiomics signature.In clinical and MRI characteristics,continuous variables were compared using two independent sample t-test or Mann-Whitney U test and categorical variables were compared using chi-square test or fisher’s exact test.Multi-factor logistic regression were used to establish personalized model that included radscore and clinical factors.The Hosmer-Lemeshow was used to test the fitting effect of the model,and draws receiver operating curve ROC to evaluate MRI feature models,radiomics models and personalized models in diagnosis efficacy in pituitary microadenoma.Delong test was used to compare the ROC curve AUC of each model.Evaluate its clinical application value through the decisioncurve.Results:1.The difference in age was statistically significant in the training group（P=0.03）,and there was no statistically significant difference in the verification group（P=0.71）.The AUC of the age in the training group and the verification group were 0.79（95%CI:0.68⁰.90）and 0.59（95%CI:0.33⁰.85）,respectively.2.After de-redundancy by m RMR,and dimensionality reduction by LASSO regression,13,15,14,13 texture features were selected from T1 WI,T2WI,enhanced T1 WI and joint sequence.According to these characteristics,the radscore of each patient is established by the linear combination of the weighting coefficient products corresponding to these characteristics.According to ROC analysis,the T1 WI radiomics model is the most effective in the identification of microprolactinomas,the AUC（95%CI）of the ROC in training and validation group was0.90（95%CI:0.83⁰.97）and 0.78（95%CI:0.60⁰.97）.The AUC（95%CI）of the ROC for T2 WI radiomics model in training and validation group was 0.88（95%CI:0.80⁰.96）and 0.77（0.59⁰.95）.The AUC（95%CI）of the ROC for T1 WI enhanced radiomics model in training and validation group was 0.75（95%CI:0.63⁰.87）and 0.72（95%CI:0.51⁰.92）.The AUC（95%CI）of the ROC for combined radiomics model in training and validation group was0.88（95%CI:0.81⁰.96）and 0.69（95%CI:0.48⁰.89）.3.The personalized model constructed by combining age and T1 WI radscore has the highest diagnostic efficiency,the AUC of the ROC in training and validation group was0.93（95%CI:0.87⁰.99）and 0.79（95%CI:0.61⁰.97）.Age and T1 WI radscore are independent risk factors for the diagnosis of pituitary microadenoma.After Hosmer-Lemeshow test,the goodness of fit of the two groups was good（p>0.05）.By Delong test,the AUC difference between the personalized model and the T1 WI radiomics model was not statistically significant（P>0.05）.The decision curve analysis shows that the personalized model has better clinical benefits,andisbetterthanthe T1 WIimageomicsmodel.Conclusion: 1.Tumor maximum diameter,semi-quantitative parameters（maximum ascending slope,peak time,maximum enhancement rate）and the signal intensity ratio of each sequence（T1WI signal intensity ratio,T2 WI signal intensity ratio and T1 WI enhanced signal intensity ratio）can not distinguish microprolactinomas and non-microprolactinomas,age has a certain rolein distinguishing.2.Among the radomics models,the T1 WI radiomics model have high diagnostic value,and werebetterthan T2 WI,T1WIenhanced,combined radiomicsmodels.3.The combination of age and T1 WI radscore to build personalized model is better than the T1 WIradiomics,andthepersonalizedmodel hasbetterclinicalbenefits.