The Value Of Radiomics Model Based On Conventional MRI Sequence In Predicting Early Recurrence Of Glioma After Surgery

Purpose:The predictive value of radiomics models based on different conventional sequences(T1WI,T2 WI,FLAIR,T1 WI enhancement)and combined sequences in patients with glioma within one year after surgery was discussed and compared.Method:From January 2015 to February 2023,40 glioma patients with complete surgical pathological confirmation were collected in the China-Japan Union Hospital of Jilin University,and these patients were divided into 20 cases in the early recurrence group and 20 cases in the non-recurrence group.All patients underwent routine MRI within1 week prior to surgery,including T1 WI,T2WI,T2-weighted FLAIR,and T1 WI enhancement.Using the 3D Slicer software,the region of interest is manually delineated at all levels of lesions on conventional MRI images of different sequences,and the software automatically fuses to generate three-dimensional ROI,and the "Pyradiomics Package" plug-in of the software is used to extract radiomic features,including morphological features(2D,3D),first-order features,texture features,filter-based(including original images,wavelet transforms),a total of 851 radiomic features.First,Mann-Whitney U was used to test the preliminary screening,then the feature correlation analysis was used,and finally the L1 regular method and the 5-fold cross-validation method were used to further screen the most valuable features,and the logistic regression model was established respectively.The receiver operating characteristic curve was used to evaluate the diagnostic performance of different sequence models.Results:The area under the curve,sensitivity,specificity and optimal cut-off points of the radiomics model constructed by the T1 WI sequence were 0.862,0.95,0.65 and 0.6,respectively;the area under the curve(AUC),sensitivity,specificity and optimal cut-off point of the radiomics model constructed by the T2 WI sequence were 0.865,0.75 and 0.95,respectively.The specificity and optimal cut-off points were 0.945,0.95,0.85 and 0.8,respectively,and the area under the curve(AUC),sensitivity,specificity and optimal cut-off points of the radiomics model constructed by T1 WI enhanced sequence were 0.993,0.95,1 and 0.95,respectively.In the multivariate logistic regression model,only the Rad-score of T1 WI and FLAIR was statistically significant,and the other sequences and other indicators were not statistically significant.Conclusion:The radiomics model constructed by conventional sequence(T1WI,T2 WI,FLAIR,T1 WI enhancement)has certain value in predicting the recurrence of glioma within one year after surgery,and the radiomics model constructed with enhanced sequence has the highest predictive efficiency,which can provide information for preoperative treatment decisions and postoperative follow-up.