Prediction Of Glioma Pathological Grading Based On Magnetic Resonance Imaging Radiomics

Objective: Accurate preoperative grading of gliomas is crucial for the treatment and assessing the prognosis of patients.To excavates the optimal characteristic parameters of MRI based radiomics on gliomas histology,exploring the application value of MRI based radiomics by grading diagnosis and improving the diagnostic efficiency of MRI in grading of gliomas.Methods: Retrospective analysis was performed on 59 cases of glioma diagnosed by pathology.All the patients underwent non-enhanced and contrast-enhanced 3.0T MRI scan before surgery.Manually delineating region of interest,and then extracting 396 high-dimensional 3D imaging ensemble features on T1 WI,T2WI,T2 FLAIR and T1 WI enhancement sequences.After data preprocessing,feature selection was performed by ANOVA,Kruskal-Wallis test and LASSO regression,through logical linearity.Multiple logistics regression and cross-validation established a predictive model for radiomics.Nomogram were established with radiomics signature and selected of clinical imaging diagnostic features including tumor hemispheres,degree of enhancement,necrosis/cysts,tumor spanning midline,and peritumoral edema.Multivariate regression analysis was used to establish the grading model.The prediction performance of the model was evaluated with ROC analysis and AUC,and the performance of the nomogram was evaluated in the terms of its calibration.Results: After feature dimension reduction,three characteristic parameters were selected from T1 WI sequence,seven characteristic parameters were selected from the contrast enhanced T1-Weighted Imaging(T1WI-CE)and six characteristic parameters were selected on T2 WI.T2 Fl AIR has no valuable features.In the T1 WI prediction model,the area under the curve(AUC)is 0.716,the sensitivity is 80.4%,and the specificity is 61.5%.After adding the clinical imaging diagnostic index,the AUC was 0.794,the sensitivity was 84.8%,and the specificity was 69.2%.The AUC in the T1WI-CE prediction model is 0.910,the sensitivity is 91.3%,and the specificity is 84.6%;After adding the clinical imaging diagnostic index,the AUC was 0.915,the sensitivity was 93.5%,and the specificity was 84.6%.In the T2 WI prediction model,the area under the curve is 0.808,the sensitivity is 58.7%,and the specificity is 92.3%;After adding the clinical imaging diagnostic index,the AUC was 0.853,the sensitivity was 92.3%,and the specificity was 69.6%.Conclusions:(1)The grading of glioma is related to the imaging features(tumor edema,necrosis/cysts,degree of enhancement).(2)Radiomics features based on T1 WI enhanced sequence(AUC=0.910)have the highest diagnosis efficacy comparing with T2 WI and T1 WI,also higher than the clinical imaging diagnostic mode(AUC=0.828).Radiomics provides an important reference for the differential diagnosis in gliomas grading.The individualized predictive model consists of tumor edema,necrosis/cyst change,degree of enhancement,and radiomics signature(based on T1WI?T2WI and T1 WI-CE respectively),individiualized predictive model is better than the diagnostic radiomics model.(3)Visual nomogram based on radiomics signature and clinical imaging diagnostic features have good clinical application prospects,that may guide more accurately grading diagnosis of glioma.