The predictive value of a nomogram model based on SWI and CE-T1WI radiomics features and clinical characteristics for IDH1 genotype in diffuse glioma

ZHAO Qirui; BAO Han; YE Junjie; LI Wenzhi; XIE Wei; WANG Qing; LING Zongfang

doi:10.20258/j.cnki.1006-9011.2025.11.002

Abstract：Objective To develop and evaluate a nomogram prediction model based on susceptibility-weighted imaging(SWI)and contrast-enhanced T1-weighted imaging(CE-T1WI)radiomics features combined with clinical characteristics for pre-dicting the isocitrate dehydrogenase1(IDH1)genotype in diffuse glioma.Methods A study was performed on 50 patients with pathologically confirmed diffuse glioma and IDH1 gene results.Two radiologists manually delineated the region of interest on SWI amplitude map and axial CE-T1WI image,respectively.Radiomics features were extracted and selected from ROI to predict IDH1 genotype of diffuse glioma.Machine learning models were trained and validated using selected radiomics features.The model with the best prediction performance was selected to calculate radiomics score(Rad-score).Then,the nomogram was built by combin-ing Rad-score with clinical factors of significant difference.Finally,the nomogram was evaluated by consistency index(C-index)and Hosmer-lemeshow(HL)test.Results NN,SVM,RF and AdaBoost prediction models of tumor area and tumor parenchy-mal area were built based on SWI,CE-T1WI,and SWI+CE-T1WI,respectively.Among the models,SWI+CE-T1WI model of tu-mor area had the best prediction performance.Based on NN,SVM,RF and AdaBoost machine learning models,the AUC values of tumor area were 0.86,0.90,0.86 and 0.75,the AUC values of tumor parenchymal area were 0.85,0.78,0.82 and 0.76,re-spectively.Among the models,SVM model of tumor area had the best prediction performance,which was regarded as the best ra-diomics prediction model.Rad-score was calculated according to the radiomics features and their coefficients.For clinical fac-tors,age was the only factor that made a statistical difference between the two groups(P<0.05).Finally,the nomogram was built to predict IDH1 genotype of gliomas by combining Rad-score with patient's age(C-index=0.965;HL test:P>0.05).Conclusions The radiomics models based on SWI,CE-T1WI and SWI+CE-T1WI have certain clinical value in predicting IDH1 genotype of diffuse glioma.The nomogram prediction model based on the Rad-score from SWI and CE-T1WI,combined with patient age,may represent a promising method for the preoperative noninvasive prediction of the IDH1 genotype in diffuse glioma.