Research On Predicting The Consistency Of Meningioma By MRI-based Radiomics

Purpose:To compare the efficacy of clinical imaging feature models,MRI-based radiomics models and combined diagnostic models in predicting the consistency of meningioma before surgery,and to explore the value of radiomics in predicting the consistency of meningioma.Methods:Retrospective analysis was executed on 200 patients with meningiomas confirmed by surgery and pathology from October 2018 to October 2020 in the First Affiliated Hospital of Wannan Medical College.According to the surgeon’s intraoperative assessment,they were divided into soft group（n=60）and hard group（n=140）.According to the tumor grade,MRI appearanc and the surgeon’s findings during the operation,they were divided into invasive group（n=96）and non-invasive group（n=104）.Patients were divided into training group（n=148）and verification group（n=52）by completely random method.All patients underwent routine head MRI scan.The ITK-SNAP software was used to manually outline the ROI layer by layer along the tumor edge on the T1WI-FLAIR,T2WI-FLAIR,and T1WI＿CE images,then three-dimensional fusion of these ROIs,the AK software was used to extract the texture feature parameters of these VOI,m RMR and LASSO regression analysis was used to analyze the texture features,perform screening and dimensionality reduction,then create imageomics tags.In clinical and conventional MRI image feature,the differences between groups of continuous variables were compared using two independent sample t-test or Mann-Whitney U test,and the differences of categorical variables were compared using chi-square test or Fisher’s exact test.Single-factor and multi-factor logistic regression were used to establish clinical imaging feature models,single-sequence and combined-sequence radiomics models,and combined diagnostic models.The reliability of these model was verified by 100-time retention cross-validation（LGOCV）.The Hosmer-Lemeshow was used to test model Fitting effect,the ROC was darwn to calculate the area under the curve,and evaluate the ability of each model to predict the consistency of meningioma,De Long was executed to compare the ROC curve and AUC of each model,to evaluate the clinical application value of the model by drawing a decision curve.Results:In the clinical features,there was no statistically significant difference in the age,gender,tumor location,and maximum tumor diameter between the soft and hard meningioma in the training group and the validation group（P<0.05）.The difference in tumor aggressiveness between the two groups was statistically significant（P-value were 0.01 and 0.03 in the training group and the validation group）.In the MRI image features,the T1WI-FLAIR,T1WI＿CE and tumor enhancement rate of the training group and the validation group were not statistically different between the different consistency groups,and T2WI-FLAIR has statistically significant differences among different consistency groups（in the training group and the validation group the P-value are all 0.04）.AK software selects 12,12,14,14 features with the most predictive value from T1WI-FLAIR,T2WI-FLAIR,T1WI＿CE and joint sequence,respectively,and obtains the rad-score of each patient by linearly combining the product of the weighting coefficients corresponding to these features.The clinical imaging feature model,T1WI-FLAIR,T2WI-FLAIR,and T1WI＿CE radiomics,combined sequence radiomics model and combined diagnosis model were established respectively,and the AUC（95%CI）of meningioma consistency was predicted in the training group and the validation group are 0.614（0.501～0.728）and 0.616（0.451～0.781）,0.751（0.661～0.841）and 0.756（0.612～0.900）,0.765（0.675～0.855）and 0.767（0.633～0.901）,0.671（0.573～0.770）and 0.737（0.588～0.887）,0.804（0.723～0.884）,0.796（0.654～0.938）and 0.814（0.734～0.895）and 0.813（0.670～0.956）.Both the combined sequence radiomics model and the combined diagnostic model have good predictive capabilities.The combined diagnostic model has the highest predictive performance.Tumor aggressiveness,R_T2-FLAIR and the combined sequence radiomics label are all independent predictors of meningioma consistency.Decision curve analysis shows that it has better clinical benefits.Conclusion:1.The consistency of meningiomas is related to the aggressiveness of the tumor,but not related to gender,age,tumor location and tumor size.2.The consistency of meningioma is related to the T2WI-FLAIR signal of MRI.Tumors with higher T2WI signal tend to be soft in consistency;the consistency of the tumor has nothing to do with the T1WI-FLAIR,T1WI＿CE of MRI and the enhancement rate of the tumor.3.Among the radiomics models,the T2WI-FLAIR sequence radiomics model predicts the consistency of meningiomas better than the T1WI sequence and the T1WI＿CE sequence;the combined sequence radiomics model is better than the three single sequence radiomics models and has better prediction value.4.Combining clinical and MRI image features and joint sequence radiomics model to jointly establish a joint diagnosis model,and the ability to predict the consistency of meningioma is further improved,and it has better clinical benefits.