Radiomics Model Building From Multiparametric MRI To Predict Grade In Patients With Glioma

Objectives: The aim of this study was to develop a radiomics nomogram using multiparametric magnetic resonance imaging(MRI)for predicting grade in patients with glioma.Methods: This study involved 85 patients(training cohort: n = 56;validation cohort: n = 29)with pathologically confirmed gliomas.Radiomic features were extracted from contrast-enhanced T1-weighted(CET1WI),axial T2-weighted(T2WI)and apparent diffusion coefficient(ADC)sequences.A least absolute shrinkage and selection operator(LASSO)regression approach was developed for dimensionality reduction,feature selection,and radiomic signature building.Radiomic signature,age,and gender were incorporated as potential predictors to perform logistic regression analysis and build a prediction model of glioma grade,and a radiomics nomogram was used to represent this model.The performance of the nomogram was assessed in terms of discrimination,calibration,and clinical utility.Significant differences in age,gender,and radiomic signature between the training cohort and validation cohort were assessed respectively using the independent samples t-test,the ? 2 test,and Mann-Whitney U test.Results: We extracted 218 features from ROIs of every sequences,and through LASSO regression,10 features were selected from the CET1 WI images,6 features from the T2 WI images,and 3 features from the ADC maps.Then,we combined three sequences and selected 15 features which showed the highest diagnostic values from654 features.The radiomic signature was significantly associated with glioma grade(p<0.001)for both the training and validation cohorts.Patients were classified into a high-risk group(rad-score>0.5039)and a low-risk group(rad-score<0.504)in accordance with the optimum cut-off value(rad-score of 0.504).The results showed good discrimination;the concordance index(C-index)was 0.963 in the training cohort and 0.951 in the validation cohort.The performance of the radiomics nomogram derived from three sequences was better than those corresponding to nomograms based on CET1 WI,T2WI or ADC alone in glioma grading.The nomogram derived from three sequences showed good discrimination;the concordance index(CI)was 0.971 in the training cohort and 0.943 in the validation cohort.The calibration curve also showed good agreement between the estimated risk and the observed risk.The decision curve demonstrated that combining three sequences had favorable clinical utility.Conclusions: In this study,we extracted large number of Radiomic features from multimodal magnetic resonance sequences and built a statistical model by these features.Compared with the traditional morphological criteria,the classification of gliomas through this statistical model can be evaluated more comprehensively,objectively and accurately.Machine learning algorithm was used to screen,reduce dimension,integrate,construct image score,and combine these features.The statistical model of nomogram has a high predictive value in glioma grading.