Application Of Radiomics Model In Predicting Radiation Encephalopathy After Radiotherapy For Nasopharyngeal Carcinoma

Objective In this study,we plan to use traditional radiomic and Delta-radiomic methods to observe the heterogeneity of brain tissue in patients with nasopharyngeal carcinoma（NPC）after radiotherapy,and evaluate a prediction model of radiation encephalopathy（REP）based on the selected features.Finally,we added clinical indicators to evaluate a radiomic plus clinical comprehensive prediction model,in order to provide a basis for predicting the potential REP in patients with NPC after radiotherapy in advance.Methods A total of 121 NPC patients who had received radiotherapy were retrospectively enrolled according to the inclusion and exclusion criteria,including 57 patients with REP and 64 patients without REP.The MRI image data and clinical information of the two groups before and after radiotherapy were collected.We used ITK-SNAP software for three-dimensional manual segmentation of the left temporal lobe in both groups of patients.Then,the region of interest（ROI）imaging omics feature was extracted by using the artificial intelligence kit（A.K）software.Delta-radiomic feature was defined the change in radiomic feature from before radiotherapy to after radiotherapy radiomic feature.The specific calculation process is that the features after radiotherapy subtract the features before radiotherapy to obtain the absolute variation of features,namely Delta-radiomic feature.the optimal features were selected using univariate logistic analysis,Pearson correlation analysis and gradient boosting decision tree（GBDT）method.Finally,the corresponding logistic regression model was evaluated according to the selected features,the clinical indicators with statistical differences were added to the radiomic model,and a comprehensive prediction model was constructed.Finally,we used Delong test to compare the predictive efficacy of several models.Results 1702 features were extracted from MRI images before and after radiotherapy,respectively,and 1702 Delta-radiomic features were obtained by formula calculation.After feature selection,3,8,and 9 effective features were retained from three modalities:contrast enhanced T1 plus T2 sequence before radiotherapy,contrast enhanced T1 plus T2 sequence after radiotherapy,and Delta contrast-enhanced T1 plus Delta-T2 sequence.Conventional radiomic model before radiotherapy yielded AUCs of 0.916 and 0.891 in the training and testing sets.Conventional radiomic model after radiotherapy yielded AUCs of 0.853 and 0.729 in the training and testing sets.AUCs of pos-radiotherapy conventional radiomic plus clinical were 0.962 and 0.894 in the training and testing sets.The model of Delta contrast-enhanced T1 plus Delta-T2 had an AUC of 0.885 in the training set and an AUC of 0.741 in the testing set,and the combined Delta-radiomicplus clinical model yielded AUCs of 0.95 and 0.903 in the training sets and the testing sets.The results of Delong test showed all the combined radiomic-clinical models predicted better than the corresponding radiomic models.Conclusion The temporal lobe radiomic biomarkers can provide a basis for predicting the occurrence of REP in patients with NPC after radiotherapy.The conventional radiomic model and Delta-radiomic model constructed based on MRI images before and after radiotherapy have good predictive value for REP.The combined model with clinical indicators and radiomics features is superior to the corresponding simple radiomics model,which is helpful for early prediction of radiation encephalopathy and timely formulation of reasonable individualized treatment plan to improve the quality of life of patients.