The Value Of Radiomics Combined With Clinical In Predicting High Proliferation Index Of Ki-67 And Postoperative Recurrence In Non-functional Pituitary Macroadenomas

The first part The value of radiomics combined with clinical in predicting High proliferation index of Ki-67 in non-functional pituitary macroadenomasObjective: Ki-67 proliferation index is usually a index used to evaluate the proliferative activity of tumor cells in clinical practice.It refers to the percentage of immunohistochemistry positive cells in all tumor cells.Ki-67 is not express in normal human pituitary tissue,but it can significantly express in pituitary adenomas.Usually,Ki-67 index ≥ 3% is considered as high proliferation,while Ki-67 index < 3% is considered low proliferation.Preoperative prediction of the Ki-67 index is valuable and may influence the surgical approach and postoperative management of patients.Pituitary tumors with high Ki-67 index require neurosurgeons to perform total resection of the tumor as much as possible and prepare for a higher probability of compression and tissue adhesion in the peritumoral structures.However,it is difficult to distinguish the proliferation index of Ki-67 on conventional images.Radiomics has higher information utilization,and its quantitative analysis method makes the results more objective.Therefore,we hope that the radiomics method can be used to combine the clinical characteristics of patients to develop a preoperative method to predict the high proliferation index of Ki-67 in non-functional pituitary macroadenomas(NFMA).Materials and methods: A total of 396 patients with NFMA who underwent pituitary tumor resection in our hospital from January 2012 to July 2022 were randomly assigned to the training set and the validation set in a ratio of 7:3,with 276 cases in the training set and 120 cases in the validation set.This study also collected 134 patients with NFMA from other hospitals from June 2012 to December 2019 as an external test set.Patient’s clinical information(age,sex,main symptoms),magnetic resonance imaging(MRI)characteristics(tumor morphology,suprasellar involvement,sphenoid sinus involvement or cavernous sinus involvement,maximum diameter of tumor,hemorrhage,size of tumor capsule,tumor signal uniformity,tumor volume,T1 WI,T2WI and T1-enhanced signal intensity ratio of the solid part,and Knosp grade),A Logistic regression analysis was used to determine the independent risk factors predicting the high proliferation index of Ki-67,and ultimately establish a clinical prediction model.MR T1 WI,T2WI,T1 enhanced images were segmented using 3D Sllicer software,and the Radiomics plug-in was used to extract the imaging features;Inter-class and intra-class correlation coefficient(ICC),analysis of variance,minimum absolute convergence and selection operator(LASSO)and logistic regression were used to filter the extracted features;Multivariate Logistics regression was used to establish the radiomics model.Through multi-factor logistic regression,using clinical and radiomics features,a combined prediction model was established.A nomogram was drawn based on the combined model and a calibration curve was used to evaluate the similarity between the predicted and actual values of the nomogram.The receiver operating characteristic(ROC)curve and the area under the ROC curve(AUC)were used to evaluate the predictive efficacy of the three models we developed.The clinical utility of the models were assessed using a decision curve analysis(DCA).The validation set and the external test set were used for the validation against the different prediction models.Results: The predictor included in the clinical model was patient age,with the AUC0.661(95%CI: 0.602-0.717)in the training set;AUC 0.551(0.457-0.642)in the validation set;and AUC 0.518(0.430-0.605)in the external test set.Through ICC,analysis of variance,LASSO dimension reduction and logistic regression,8 radiomics features were selected,and were used to establish the radiomics model.The AUC of the model in the training set was 0.861(95% CI: 0.814-0.899);the AUC in the validation set is 0.848(95% CI: 0.771-0.907);the external test set is 0.793(95% CI: 0.715-0.858).A combined prediction model was established by combining the clinical features and the radiomics features.The AUC of the model was 0.875(95% CI: 0.830-0.911)in the training set;the AUC was 0.871(95% CI: 0.798-0.926)in the validation set;the external test set is 0.804(95% CI: 0.726-0.867).The Delong test showed that the combined model had the same efficacy as the radiomics model,which were higher than the clinical model.The calibration curve indicated a high similarity between the predicted values and the actual values of the combined model.The DCA curve showed that the combined model and the radiomics model had high returns in most of the threshold probability range,and the net benefit of the clinical model was extremely low.Conclusion: The clinical prediction model constructed based on clinical features alone has insufficient prediction efficacy and is of little value in predicting Ki-67 high proliferation index in patients with NFMA.The radiomics model and combination model have high diagnostic value in predicting Ki-67 high proliferation index in patients with NFMA before surgery,which is beneficial to guide the choice of clinical treatment options.The second part The value of radiomics combined with clinical in predicting postoperative recurrence in non-functional pituitary macroadenomasObjective: Although the majority of nonfunctional pituitary macroadenomas(NFMA)are diagnosed as benign adenomas,suprasellar or parassellar invading NFMA are difficult to be removed completely.In 12%-58% of patients with residual adenomas will relapse.Even if the adenoma is completely removed,10%-20% of the tumors will recur within 5-10 years.Although postoperative adjuvant radiotherapy can be used to reduce the recurrence of NFMA,this approach may lead to irreversible pituitary insufficiency and other long-term complications.Therefore,it is important to investigate the predictive factors of recurrence of NFMA.Previous studies have shown that many clinical,pathological,and traditional MRI features are associated with postoperative recurrence of NFMA.However,most of these parameters are qualitative and subjective,with inter-observer differences and low predictive efficacy.Radiomics is an analysis form that quantitatively extracts imaging features from medical data,its information utilization is higher,and its quantitative analysis method makes the results more objective.Therefore,we hope to establish a comprehensive classification model,combining clinicopathological risk factors and preoperative radiomics characteristics,to predict recurrence of NFMA within 5 years after surgery and thus providing an effective tool for the choice of clinical treatment options.Materials and methods: A total of 292 patients with NFMA who underwent pituitary adenoma resection in our hospital from January 2012 to January 2018 were randomly assigned to the training set and the verification set in a ratio of 7:3,with 204 cases in the training set and 88 cases in the validation set.This study also collected 123 patients with NFMA from other hospitals from June 2012 to January 2018 as an external test set.The patient’s basic clinical information(age,gender,clinical symptoms),surgical and postoperative pathological information(tumor resection degree,tumor texture,tumor hormone type,Ki-67 proliferation index),conventional imaging features(tumor morphology,invasion site,maximum diameter,volume,hemorrhage,cyst,uniformity of tumor signal,Knosp grade of tumor,signal intensity ratio of T1 WI,T2WI and T1 enhancement of solid part),were analyzed by univariate and multivariate logistic regression analysis,to identify the independent risk factors for predicting the recurrence of NFMA after surgery.Finally,the clinical prediction model was established,and the nomogram was drawn according to the clinical model,and the calibration curve was used to evaluate the similarity between the predicted and actual values of the clinical model.MR T1 WI,T2WI,and T1 enhanced images were segmented using 3D Sllicer software,and radiomics features were extracted using Radiomics plug-in;ICC,ANOVA,LASSO and logistic regression were used to screen the extracted features;Multi-factor Logistic regression was used to establish the radiomics model.Through multivariate logistic regression,the combined prediction model was established using clinical features and Radiomics features.The nomograph was drawn according to the combined model and the calibration curve is used to evaluate the similarity between the predicted value and the actual value.The ROC curve and AUC value are used to evaluate the predicted effectiveness of the three models.Decision curve analysis(DCA)was used to evaluate the clinical applicability of the model.Finally,the different prediction models are verified in the validation set and the external test set.Patients were divided into high-risk group and low-risk group using the best cutoff value of the combined model,and Kaplan-Meier survival analysis was performed.Used Cox regression to construct a combined predictive model based on patient’s survival time,clinical,pathological,conventional imaging characteristics and Radiomics features.Draw a nomograph of 1,3,and 5 years of recurrence free survival(RFS).Predictive efficacy was evaluated using the consistency index(C-index),and its calibration curve was plotted.Results: The independent risk factors selected by clinical models included:dumbbell shape,irregular shape and lobulated shape,subtotal resection and major resection,PRL positive,Ki-67≥3,T2 signal intensity ratio.The AUC of the model in the training set is 0.834(95% CI: 0.775-0.882);the AUC in the validation set is 0.756(95%CI: 0.653-0.841);in the external test set is 0.757(95% CI: 0.671-0.830).Through ICC,analysis of variance,LASSO dimensionality reduction and logistic regression,a total of14 radiomics features were selected,to establish the radiomics model.The AUC of the model in the training set was 0.88(95% CI: 0.827-0.921);in the validation set is 0.841(95% CI: 0.748-0.911);external test set is 0.835(95% CI: 0.757-0.896).A combined prediction model was established by combining clinical features with radiomics features,whose AUC was 0.908(95% CI: 0.860-0.944)in the training set;the AUC in the validation set is 0.888(95% CI: 0.803-0.945);in the external test set is 0.863(95% CI:0.790-0.919).A Delong-test of the predictive efficacy of the different models revealed that the combined model was optimal with no difference between the clinical and radiomics model.The DCA curve showed that in the majority of the threshold probability range,the net benefit of the clinical model,radiomics model and combined model were good,with the combined model being the best.Drawing normogram from the clinical model,the correction curve indicates a high similarity between the predicted and actual values.Drawing nomograms from the combined model,the correction curve indicates a high similarity between the predicted and actual values.Patients were divided into the low and high risk groups based on the best cut-off value of the combined model,there was a statistical difference in the relapse-free survival rate between the two groups.Cox regression was used to establish a nomogram for predicting RFS,The C-index was0.816(95% CI: 0.792-0.839)in the training set,0.796(95% CI: 0.760-0.833)in the validation set and 0.786(95% CI: 0.753-0.819)in the external test set.The calibration curve shows that the predicted effect of the nomogram is good.Conclusion: Both clinical model and radiomics model can help predict tumor recurrence within 5 years in patients with NFMA after surgery.The combined model predicted better than the clinical model and radiomics model.The nomograph established by Cox regression can well predict RFS of patients with NFMA.This will help predict the recurrence of patients with NFMA within 5 years after surgery,guiding the choice of postoperative follow-up strategies for patients with NFMA and guiding early adjuvant therapy in high-risk patients.