The IDH Genotype Prediction In Glioma With Radiomics Based On Magnetic Resonance Quantitative Susceptibility Mapping

Objective:To explore the value of Radiomics based on Quantitative susceptibility mapping(QSM)for predicting isocitrate dehydrogenase(IDH)genotype in glioma.Methods:A total of 45 patients with pathological confirmed glioma were retrospectively analyzed in our hospital between November 2020 and November 2022 according to the inclusion and exclusion criteria.The DICOM images of conventional magnetic resonance imaging and QSM,clinical data,and molecular marker results of the research subjects were collected.The original QSM images were processed on GE Advanced Workstation 4.4 to obtain QSM images.Regions of interest about tumor were manually segmented using ITK-SNAP software and radiomics features were extracted using the open-source software FAE.A total of 1780 features were extracted from each patient’s CE-T1WI、T2FLAIR and QSM images respectively.Z-score,least absolute shrinkage and selection operator(LASSO)regression analysis、Pearson correlation coefficient、recursive feature elimination and single factor analysis were used to screen features and build a radiomics model.Four groups of radiomics models(CE-T1 WI group,T2 FLAIR group,QSM group,combined sequence group)were established by linear fitting according to the weight of feature coefficient.The consistency index(C-index)of each radiomics model was calculated to obtain the best radiomics model.The Delong test was used to compare the AUC values between different models.In addition,the generalization of the combined model was explained using leave-one-out cross validation.Results:We screened 4,4,4 and 5 radiomics features from CE-T1 WI,T2FLAIR,QSM and combined sequences,respectively.Four radiomics prediction models were constructed based on the selected features,Radscore＿CE-T1 WI,Radscore＿T2FLAIR,Radscore＿QSM,and Radscore＿com.The AUC value of each prediction model were 0.812、0.848、0.922、0.947,respectively.The C-index of each prediction model were 0.740(95% CI: 0.590-0.890)、0.736(95% CI: 0.587-0.885)、0.830(95% CI: 0.696-0.964)、0.872(95% CI: 0.662-0.922),respectively.Conclusion:This study screened out the imaging biomarker for predicting glioma IDH genotyping based on QSM radiomics.The AUC value of QSM prediction model is the highest in a single sequence,which indicates that the prediction ability of QSM sequence is better than other sequences.In addition,the AUC value of the QSM combined with the conventional sequence prediction model was further improved,which was the best radiomics model for predicting glioma IDH genotyping,indicating that the predictive efficiency of the radiomics model constructed by the combined sequence was superior to the single sequence.Therefore,the QSM-based radiomics model can predict the IDH genotyping of glioma and guide the formulation of clinical treatment plans.