MRI Radiomics Predicts Short-term Efficacy And Recurrence Risk Of Locally Advanced Nasopharyngeal Carcinoma

Part I: MRI radiomics predicts the efficacy of induction chemotherapy for locally advanced nasopharyngeal carcinoma.Objective: To explore the value of MRI radiomics in prediction for the efficacy of induction chemotherapy in patients with locally advanced nasopharyngeal carcinoma.Method: The clinical and imaging data of 257 patients with stage III-IVa nasopharyngeal carcinoma confirmed by pathological biopsy and treated with IC were retrospectively analyzed.All patients underwent magnetic resonance imaging before and after IC treatment,and were divided into sensitive group and resistant group according to RECIST.Ma Zda software was used to manually delineate the ROI of the largest level of the lesion on the axial T2 WI and axial CET1 WI images of the magnetic resonance before treatment,and the image feature parameters of the nasopharyngeal carcinoma lesion were extracted.The feature selection method provided by the software: Fisher coefficient,interaction information(MI),classification error probability combined with average correlation coefficient(POE +ACC)combined to screen the best texture features.The initial data analysis(RDA),principal component analysis(PCA),linear discriminant analysis(LDA)and nonlinear discriminant analysis(NDA)of B11 statistical module are used to classify and analyze the selected texture features.LASSO regression was used to screen the better feature parameters,and 257 sample data were divided into validation set and test set according to the ratio of 7: 3.Based on the machine learning KNN algorithm,the radiomics models of T2 WI,CET1WI and their combination were established respectively,and the ROC curve was drawn by logistic regression to calculate the AUC value to evaluate the prediction efficiency of the model.Result: 9 features were selected from T2 WI.The T2WI-based radiomics signature yielded the AUC of The AUC of 0.590 [ 95 % CI(0.463 ～ 0.717)],the sensitivity of 53.8 %,and the specificity of 64.1 %.16 features were selected from the CET1 WI.The CET1WI-based radiomics signature yielded the AUC of 0.633[ 95 % CI(0.480-0.786)],the sensitivity of 61.5 %,and the specificity of 65.4 %.24 features were screened from the multiple images.The multiple images-based radiomics model yielded the AUC of 0.815 [ 95 % CI(0.726-0.904)],the sensitivity of 86.7 %,and the specificity of63.2 %.The multiple images-based radiomics model shows the better prediction performance.Conclusion: The radiomics features based on T2 WI and CET1 WI are related to the efficacy of LANPC induction chemotherapy.The radiomics model combined with T2 WI and CET1 WI can predict the efficacy of induction chemotherapy in patients with locally advanced nasopharyngeal carcinoma,which can help clinical patients with individualized precision diagnosis and treatment.Part Ⅱ:MRI radiomics to evaluate the risk of recurrence after radiotherapy and chemotherapy in patients with locally advanced nasopharyngeal carcinomaObjective : To explore the application value of MRI radiomics in the risk assessment of recurrence after radiotherapy and chemotherapy in patients with locally advanced nasopharyngeal carcinoma.Method: The clinical and imaging data of 83 patients with stage III-IVa nasopharyngeal carcinoma confirmed by pathological biopsy and receiving wholecourse radiotherapy and chemotherapy were retrospectively analyzed.All patients underwent magnetic resonance examination before radiotherapy and chemotherapy,and were followed up during radiotherapy and chemotherapy and after the end of treatment.The follow-up time was 36-60 months.According to evaluation of nasopharyngeal and systemic conditions by imaging examination,clinical diagnosis and pathological diagnosis after treatment,the patients were divided into two groups,recurrence group(n = 40)and non-recurrence group(n = 43).The clinical data were analyzed by independent sample T test and multivariate Logistic regression.Ma Zda software was used to manually delineate the region of interest(ROI)at the largest level of the lesion on the axial T2 WI and axial T1 WI enhanced scan(CET1WI)images before treatment,and the radiomics features of nasopharyngeal carcinoma lesions were extracted.Fisher coefficient,interaction information(MI),classification error probability combined with average correlation coefficient(POE + ACC)were used to screen the best texture features.The initial data analysis(RDA),principal component analysis(PCA),linear discriminant analysis(LDA)and nonlinear discriminant analysis(NDA)of B11 statistical module are used to classify and analyze the selected texture feature parameters.Then,LASSO regression is used to further reduce the dimension of the feature parameters,and the sample data is divided into a verification set and a test set according to the ratio of 7 : 3.Based on the decision tree(DT)machine learning method,the radiomics models of T2 WI,CET1WI and their combination were established respectively,and the receiver operating characteristic curve(ROC)was drawn to calculate the area under the curve(AUC).Result: 12 features were selected from T2 WI.The T2WI-based radiomics signature yielded the AUC of The AUC of 0.580 [ 95 % CI(0.351 ~ 0.809)],the sensitivity of 50.0%,and the specificity of 67.9 %.17 features were selected from the CET1 WI.The CET1WI-based radiomics signature yielded the AUC of 0.587 [ 95 % CI(0.359-0.814)],the sensitivity of 25.0 %,and the specificity of 76.9 %.9 features were screened from the multiple images.The multiple images-based radiomics model yielded the AUC of 0.731 [ 95 % CI(0.517-0.887)],the sensitivity of 75.0 %,and the specificity of 61.5 %.The multiple images-based radiomics model shows the better prediction performance.Conclusion: The radiomics features based on T2 WI and CET1 WI images are associated with the risk of recurrence after LANPC chemoradiotherapy.The radiomics model combined with T2 WI and CET1 WI features can predict the risk of recurrence in patients with locally advanced nasopharyngeal carcinoma.