Prediction Of P53 Status In Glioma By Radiogenomics Model Based On Diffusion Tensor Imaging

Objective:To investigate the predictive effect of radiogenomics model based on diffusion tensor imaging(DTI)on p53 expression in gliomas.Methods:Sixty-four patients with pathologically confirmed gliomas in the First Affiliated Hospital of China Medical University from August 2014 to January 2020 were retrospectively analyzed,including 30 cases of p53 mutant type and 34 cases of p53 wild type.All patients underwent routine and diffusion tensor imaging before operation.The DICOM images of each sequence were registered after format conversion by MRIcron software,and the registered images were resampled to 5×5×5 mm~3on A.K software.In ITK-SNAP software,the abnormal high signal area including tumor parenchyma and peritumoral edema was delineated layer by layer on T2WI images,namely the region of interest,and the ROI of each layer formed the volume of interest.The VOI in T2WI was copied to ADC maps and FA maps.A.K software was used for feature extraction,and Spearman correlation analysis and LASSO algorithm were used for feature dimensionality reduction to establish Logistic regression model.The predictive efficacy of the model for the expression state of p53 in glioma was evaluated by the receiver operating characteristic curves.The differences of age,gender,WHO classification and other clinical data and the selected characteristics of each model between mutant p53 group and wild-type p53 group were statistically analyzed by independent sample t test,chi square test or Mann Whitney U test,and Bonferroni correction was applied to the multiple hypothesis test.Double tail p value<0.05 showed that the difference was statistically significant.Results:there was no significant difference in age,gender and glioma WHO grade between mut p53 group and wt p53 group.391 features were extracted from each MRI sequence.After feature screening,11,14 and 10 features were selected from T2WI,ADC and FA images respectively.The selected features of each sequence were compared between mut p53 group and WT p53 group.In T2WI,there was no significant difference in the selected features;in ADC maps,there were two features with statistical difference(P<0.05),that is,Cluster Shade＿All Direction＿offset1＿SD and High Grey Level Run Emphasis＿All Direction＿offset1＿SD;in FA maps,there were four characteristics with statistical difference(P<0.05),namely,variance,kurtosis,GLCM entropy＿All Direction＿offset7＿SD and Short Run Emphasis＿angle45＿offset4.After correction by Bonferroni,only one feature in the FA maps showed statistical difference between the mut p53 group and the wt p53 group,namely,GLCMEntropy＿All Direction＿offset7＿SD(p<0.005).The prediction performance of each model was evaluated by ROC curve.In the training set,the AUC values of the radiogenomics models constructed by T2WI,ADC and FA were 0.816,0.834 and 0.822,respectively.In the testing set,the AUC values of the radiogenomics models constructed by the T2WI,ADC and FA were 0.747,0.758 and 0.798,respectively.Conclusion:This study indicates that radiogenomics model based on diffusion tensor imaging may be effective and non-invasive for preliminary prediction of p53 expression status in glioma,which is expected to provide certain reference value for the application of chemotherapy drug temozolomide(TMZ)in clinical glioma patients.