To Explore The Feasibility Of Pre-Academic Prediction Of Risk Grade And Ki-67 Expression Level Of Gastrointestinal Stromal Tumor Based On MRI Imaging Group

Part One:Correlation between MRI imaging histological features and risk classification of gastrointestinal mesenchymal tumors and Ki-67 indexObjective: To analyze the correlation between ADC imaging characteristic parameters and the grade of pathological risk and KI-67 proliferation index of Gastrointestinal stromal tumors(GISTs),to explore the feasibility of correlation characteristic parameters in predicting the preoperative pathological risk classification of GISTs.Materials and methods: From January 2012 to December 2021,50 patients with GISTs who underwent preoperative MRI scan and were confirmed by postoperative pathology in our hospital were retrospectively analyzed.Gastrointestinal stromal tumors(GIST)were divided into three groups according to the NIH standard: very low-risk group(N = 21),medium-risk group(N = 16)and high-risk group(N = 13).Regions of interest(ROI)were delineated on the ADC images of the above three groups of gastrointestinal stromal tumors with different risk grades.The characteristic parameters were screened and analyzed by using ANOVA,K r u s k a l-W a l l i s Test and S p e a r m a n correlation analysis.The diagnostic efficacy of various characteristic parameters to identify GISTs with different pathological risk grades was evaluated by receiver operating characteristic(ROC)curve,and the characteristic parameters that can be used to independently predict tumor risk were selected from the characteristic parameters with statistically significant difference between groups by logistic regression analysis.Results: There were significant differences in the differential diagnosis of gastrointestinal stromal tumors between the very low-low risk group and the high risk group with Long Run Emphasis＿angle0＿offset4,Long Run Emphasis＿angle45＿offset1,and Long Run Emphasis＿angle90＿offset4(P < 0.05).In the differential diagnosis of gastrointestinal stromal tumors between very low-low risk group and high risk group,and between intermediate risk group and high risk group,uniformity and sum Average were statistically significant(P < 0.05).The areas under the ROC curve were 0.765,0.716,0.768,0.775 and0.826 respectively;After fitting 5 texture feature parameters,the area under ROC curve was0.856.S p e a r s o n analysis showed that the above five characteristic parameters were significantly correlated with tumor risk grade and Ki-67 index.Conclusion: It is valuable to predict the pathological risk grade and KI-67 index of gastrointestinal stromal tumors(GIST)based on imaging parameters,which has a certain positive significance for guiding clinical treatment and evaluating prognosis of patients.Part Two:Feasibility study on predicting Ki-67 expression levels in gastrointestinal mesenchymal tumours based on an MRI imagingomics modelObjective:To investigate the feasibility of preoperative non-invasive prediction of KI-67 expression levels in gastrointestinal mesenchymal tumours based on magnetic resonance imaging histology models.Materials and Methods: Retrospective analysis of 83 patients in our hospital who received MRI scan preoperatively and pathologically proven after surgery from January 2012 to January 2023,the GISTs are classified into a high ki-67 PI group(ki-67 PI ≥ 5%,n= 49)and low ki-67 PI group(ki-67 PI < 5%,n= 34).According to the 7:3 random stratified sampling principle,the sample data in this study were divided into test set(n=58)and verification set(n=25).Respectively for the two groups(high ki-67 PI,low ki-67PI)in patients with mri ADC(ROI)image outlined areas of interest,will split the tumor volume figure(ROI)import number of scientific research platform automatic feature extraction of omics,In addition,1874 feature parameters were standardized by the feature screening module of the platform,and then features with absolute value of Pearson correlation coefficient ≥0.9 were removed by Spearman correlation analysis,univariate screening(80percentile was selected),and statistically significant feature parameters were screened by Lasso dimension reduction method.Then,logistic regression(LR),random forest(RF),support vector machine(SVM),Stochastic Gradient Descent(SGD),Support vector machine Stochastic Gradient Descent and linear Support vector machine classification(LINEARSVC)were established.Accuracy,sensitivity,specificity,and area under subject operating characteristic curve(AUC)were used to compare the predictive efficacy of the models,and model calibration curves were constructed to evaluate the models.Decision curve analysis was used to evaluate clinical benefits.Result:Three characteristic parameters were screened by correlation analysis,Univariate filtering(Selected percentile 80)and Lasso dimensionality reduction method,They are auto__wavelet-HLH_glszm_Zone Entropy,auto__wavelet-LLH_firstorder_Kurtosis,and auto__wavelet-LLH_glrlm_LongRunEmphasis.There was no significant difference in Ki-67 expression among different sex and age groups in GIST patients(X2=1.096,P=0.295,P>0.05 and X2=0.001,P=0.974,P>0.05).MRI imaging features,tumor boundary,morphology,necrosis,calcification,enhancement pattern,growth pattern and growth location were not correlated with ki-67 expression ability in gastrointestinal stromal tumors,but tumor size was significantly different between high Ki-67 PI group and low Ki-67 PI group(p<0.001).Among the five machine learning models,Logistic Regression model has the best performance,with AUC of 0.892 and 0.74(95%CI of 0.805-0.979 and 0.534-0.946),sensitivity of 0.839 and 0.643,specificity of 0.846 and 0.727,respectively,in test set and verification set.The accuracy rates were0.842and0.68,positive predictive values were 0.867 and 0.75,negative predictive values were 0.815 and 0.615,false positive rates were 0.154 and0.273,and false negative rates were 0.161 and 0.357,respectively.Conclusions: This study shows that the imaging omics model constructed based on magnetic resonance ADC images can effectively predict the level of Ki-67 expression in gastrointestinal stromal tumor tissues,and also provides a new noninvasive prediction method for evaluating the level of Ki-67 proliferation index in gastrointestinal stromal tumor cells.