Radiomics Based On Diffusion Kurtosis Imaging To Predict The Pathological Grade Of Meningioma

Objective:To explore the predictive value of radiomics based on diffusion kurtosis imaging(DKI)for the pathological grade of meningioma,and to construct a predictive model for the pathological grade of meningioma,so as to provide help for patients and clinicians to choose treatment methods and evaluate prognosis.Methods:This study retrospectively analyzed 50 patients with meningioma confirmed by pathology in our hospital from November 2018 to December 2022.Among them,41 patients with WHO grade 1 meningioma were included in the low-level group and 9 patients with WHO grade 2 meningioma were included in the high-level group.The clinical data,pathological results,conventional MRI sequences(including T1 WI,T2WI,T2-FLAIR,CET1WI)and DKI sequence DICOM images of the subjects were collected.Based on the conventional MRI sequence images of the patients,the conventional imaging features of the tumors(including location,whether there was dural tail sign,whether the shape was regular,whether the boundary was blurred,whether the enhancement was uniform,tumor volume and maximum radius of peritumoral edema)were evaluated and recorded.Statistical analysis was performed using IBM SPSS Statistics 29.0 to compare whether the conventional imaging features between the two groups were statistically different.Mean kurtosis(MK)and mean diffusivity(MD)parameter maps were obtained by post-processing the original image of DKI sequence.Two radiologists used 3D-slicer software to draw tumor regions of interest on CE-T1 WI,T2-FLAIR,MK and MD images,respectively.FAE software was used to extract radiomics features.Interclass correlation coefficient(ICC)was used to preliminarily screen out stable radiomics features.The least absolute shrinkage and selection operator(LASSO)was used to further screen the radiomics features most related to the pathological grade of meningioma.The conventional MRI sequence model was constructed based on CE-T1 WI and T2-FLAIR radiomics features,the DKI sequence model was constructed based on MK and MD radiomics features,and the combined sequence model was constructed by combining the radiomics features of the above four sequences.After internal validation of the above three models by Bootstrap method(repeated 500 times),the area under curve(AUC),sensitivity and specificity of the receiver operating characteristic(ROC)curve analysis model were plotted,and the clinical decision curve was drawn to further evaluate the optimal radiomics model.Delong test was used to compare the differences in AUC among the above three groups,and P < 0.05 was considered statistically significant.Results:There was no significant difference in gender,age and meningeal tail sign between the low-grade group and the high-grade group(P>0.05),while there were significant differences in tumor location,irregular shape,blurred boundary,uneven enhancement,tumor volume and maximum radius of peritumoral edema(P< 0.05).Four,four and five radiomics features most related to the pathological grade of meningioma were screened by radiomics method to establish conventional MRI sequence model,DKI sequence model and combined sequence model.After internal verification by Bootstrap method,ROC curve analysis showed that the AUC of conventional MRI sequence model in predicting the pathological grade of meningioma was 0.711(95 % CI: 0.7009-0.7202),the sensitivity was 65.7 %,and the accuracy was 85.7 %.The AUC of the DKI sequence model in predicting the pathological grade of meningioma was 0.864(95 % CI: 0.8589-0.8684),the sensitivity was 91.3 %,and the accuracy was 69.4 %.The Delong test results showed that the difference in AUC between the conventional MRI sequence model and the DKI sequence model was statistically significant(P < 0.01),and the latter was superior to the former in predicting the pathological grade of meningioma.The AUC of the combined sequence model in predicting the pathological grade of meningiomas was 0.869(95% CI: 0.8642-0.8748),with a sensitivity of 81.7% and an accuracy of 85.3%,significantly improving the prediction performance compared to conventional MRI sequence models.The results of Delong test showed that there were significant differences in AUC between conventional MRI sequence model and combined sequence model(P < 0.05),DKI sequence model and combined sequence model(P < 0.05).Conclusion:The DKI sequence model has good predictive performance for the pathological grade of meningioma,which is superior to the conventional MRI sequence model.The predictive performance of the combined sequence model for the pathological grade of meningioma was also significantly higher than that of the conventional MRI sequence model,which proved the value of DKI-based radiomics in predicting the pathological grade of meningioma and was expected to assist the clinical management of meningioma patients before surgery.