Qualitative,Quantitative And Radiomics Analysis Of Multimodal MRI In Grading Brain Gliomas And Predicting Proliferation

Objective:Glioma is the most common primary brain tumor in adults and has a poor prognosis.Accurate preoperative classification is an important prerequisite for individualized treatment and the guarantee of the best prognosis.MR is the main method for preoperative evaluation of glioma.This study discussed the application value of qualitative analysis based on conventional MR sequence,quantitative analysis based on multimodal MR sequence,and multimodal MR image Radiomics analysis based on 3D segmentation technology in preoperative pathological grading diagnosis of glioma,as well as its ability to predict tumor proliferation activity.Methods:According to the inclusion and exclusion criteria,patients with suspected glioma were subjected to routine enhanced MR examination and functional MR examination.The differences in clinical-imaging characteristics of patients with different grades of glioma were compared,including gender,age,tumor location,size,nature,peritumor edema degree,and enhancement mode,as well as the correlation between various characteristics and ki-67.The diagnostic efficacy of the qualitative analysis based on conventional MR sequence in predicting the pathological grading of glioma and its correlation with ki-67 were evaluated.Functional MR sequences were quantitatively analyzed,and tumor parenchyma was selected for regions of interest(ROI).The difference of each quantitative parameter between different grades of glioma was compared,and the ROC curve analysis of each quantitative parameter was carried out.Logistic stepwise regression analysis was used to perform stepwise screening of significant quantitative parameters,fit the grading diagnosis model for glioma of high and low grade.ROC curve analysis was performed on the model,and the difference in the diagnostic efficacy of multimodal MR quantitative model and qualitative analysis for grading diagnosis of glioma of high and low grade was compared.All subjects from two centers were randomly divided into the training set and the test set according to the ratio of 7:3,and the differences of basic information between the training set and the test set were compared.Manual 3D full-tumor image segmentation was performed on pre-and post-contrast T1 WI,T2WI,ADC and FA images of all cases,and full-tumor image segmentation was performed on the 2D single slice of the tumor.ROI image radiomics features were extracted by AK software.The basic test(parametric test,non-parametric test,correlation analysis)and LASSO regression were used to reduce and screen the extracted texture features.Logistic linear regression was used in combination with the ten-fold cross validation to fit the image radiomics model of grading of glioma in the training set.The best radiomics characteristics of grading diagnosis of each sequence were obtained.The sensitivity,specificity and area under the ROC curve(AUC)of each sequence model applied alone were compared.The optimal texture features selected from each sequence were combined to establish a joint model and ROC curve analysis was conducted.The individual and joint models obtained from the training set were applied to the test set to verify the stability of the model.The metabolite ratios of Cho/Cr,NAA/Cr,Cho/NAA and DTI were measured to obtain the optimal grading diagnostic quantitative parameters in each functional MR sequence.According to AIC criteria,the best model including optimal radiomics feature model,tumor quantitative parameters and clinical-image label was established,and the nomogram was developed to achieve model visualization.Results:A total of 68 patients were included in the qualitative and quantitative analysis,among the basic features,peritumoral edema showed statistically significant differences between Grade ? and Grade ?(p = 0.037),Grade ? and Grade ?(p < 0.0001),high and low grade gliomas.Enhancement showed statistically significant differences between Grade ? and Grade ?(p < 0.0001),Grade ?and Grade ?(p < 0.0001),high and low grade gliomas.The glioma lesions were mainly solid(79.41%)and the remaining were cyst-solid.Tumor nature showed statistically significant differences between Grade ? and ?(p = 0.005),high and low grade gliomas.Hemorrhage occurred in the high grade glioma,including one of Grade ?(7.69%),six of Grade ?(37.5%).There was a moderate to significant correlation between peritumor edema,tumor enhancement and cystic degeneration in glioma,among which the correlation between tumor enhancement and cystic degeneration was the strongest.The AUC for grading diagnosis of glioma based on the conventional enhanced MR qualitative analysis was 0.803(95%CI,0.689-0.890,ACC,0.779,SEN,0.966,SPE,0.641).The qualitative analysis showed a significant correlation with ki-67(r=0.568,P<0.0001).Cho/Cr,NAA/Cr,Cho/NAA,MD,rASL,ITSS were statistically different between Grade? and ?,Grade ? and Grade ?,but not statistically different between Grade ? and Grade ?.The ROC curve analysis results of quantitative parameter grading diagnosis between HGG and LGG showed that the ROC curve analysis results of Cho/Cr,NAA/Cr and Cho/NAA ratios,MD values,rASL and ITSS were statistically significant and had high diagnostic efficacy.The maximum AUC of Cho/NAA was 0.866(SEN,0.862;SPE,0.821).The quantitative parameters showed a weak to significant correlation with ki-67.Logistic regression analysis performed stepwise regression analysis for the above 6quantitative parameters,and the results showed that the combined application of MD,Cho/NAA and ITSS had the highest grading diagnostic efficacy.The regression function was as follows: Logit(P)= 5.114+1.898 × ITSS-9.310 × MD+0.215 ×Cho/NAA,with a cut-off value of 0.537.If Logit(P)<0.537,a low grade glioma was diagnosed.If Logit(P)?0.537,a high-grade glioma was diagnosed.The AUC of the regression model was 0.959(95%ci,0.881-0.992)for the grading diagnosis of high and low grade glioma(SEN,0.966;SPE,0.949).The quantitative analysis showed a significant correlation with ki-67(r=0.587,P<0.0001).A total of 100 patients from two centers were included for image radiomics analysis,and there were no statistical differences in age,gender,tumor size,tumor location,pathological type and grade between the training set and the test set.The optimal grading radiomics features obtained from the dimensional reduction of each sequence in the training set were 3from pre-contrast T1 WI,6 from post-contrast T1 WI,7 from T2 WI,8 from ADC and 2from FA.A total of 26 radiomics features were included in the joint diagnosis,and a total of 14 texture features were obtained by basic test and LASSO regression dimensionality reduction.The ROC curve analysis of pre-and post-contrast T1 WI,T2WI,ADC,FA in the grading diagnosis model of glioma applied alone and in combination showed that the grading diagnosis efficiency of post-contrast T1 WI applied alone was the highest(AUC,0.924;ACC,0.871;SEN,0.889;SPE,0.853);The efficacy of joint application was superior to the separate application of each sequence(AUC,0.967;ACC,0.929;SEN,0.944;SPE,0.912).The single sequence of each training set and the joint radiomics model were applied to the test set,and ACC was relatively high.There was no statistical difference between the training set and the test set in ROC curve analysis results.The optimal diagnostic parameters of MRS and DTI sequences were Cho/NAA and MD respectively.According to AIC criterion,the model with the lowest AIC was selected,and the final nomogram included the joint radiomics model and Cho/NAA.The calibration curve showed that the model had a high diagnostic efficacy.Conclusion:1.In conventional MR,peritumoral edema degree,enhancement are different between Grade ? and Grade ?,Grade? and Grade ?,high and low grade gliomas.Cystic change are different between Grade ? and Grade ?,high and low grade gliomas.There is correlation among the three features,which can be the main features for gliomas grading with conventional MR sequences.2.The qualitative analysis based on the characteristics of conventional MR can be used to grade diagnosis of high and low grade glioma,with a medium diagnostic accuracy and a certain predictive power for the proliferation activity of tumor cells.3.The quantitative parameters of multimodal MR Cho/Cr,NAA/Cr,Cho/NAA,MD,rASL and ITSS show differences between Grade ? and ?,Grade? and Grade ?,high and low grade gliomas.Each quantitative parameter alone can be used to grade the diagnosis of high grade and low grade glioma,and the diagnostic accuracy is moderate.4.The logistic stepwise regression model with multi-modal MR quantitative parameters can effectively grade diagnosis of high and low grade glioma,with high diagnostic accuracy and superior to qualitative analysis,and has certain predictive power for the proliferation activity of tumor.5.The multi-modal MR sequences(T1WI,T2 WI,T1WI+C,ADC,FA)radiomics model based on 3D whole-tumor image segmentation can perform grading diagnosis of glioma.6.The imaging radiomics model has certain prediction ability on proliferation activity;7.3D is highly correlated with the diagnostic effectiveness of 2D single slice segmentation model,and 2D maximum layer segmentation is expected to replace 3D image segmentation.8.The nomogram,including the image radiomics signature and quantitative parameters is expected to provide an intuitive assessment tool for clinical practice.