The Evaluation Of The Efficacy Of Chemoradiotherapy Of Nasopharyngeal Carcinoma With MRI Radiomics

PART 1 Evaluation of short-term efficacy of MRI radiomics in nasopharyngeal carcinoma after radiotherapy and chemotherapyObjective: To explore whether MRI radiomics can predict the short-term efficacy of nasopharyngeal carcinoma after radiotherapy and chemotherapy.Methods: Retrospective analysis was performed on 92 patients with nasopharyngeal carcinoma(62 cases in sensitive group and 30 cases in insensitive group)who were pathologically confirmed to be nasopharyngeal carcinoma in our hospital from January2013 to October 2020.MRI plain scan and enhanced examination of nasopharynx and neck were performed within two weeks before treatment and within 2-3 months after the end of induction chemotherapy plus whole-course concurrent chemoradiotherapy,respectively.Patients were divided into sensitive group and insensitive group according to the efficacy evaluation of RECIST.ITK-Snap(version 3.6)was used by two radiologists to sketch the region of interest(ROI)of the largest level of the lesion on the pre-treatment MRIT1 WI,T2WI and T1 WI enhancement DICOM images,and intraclass correlation efficient(ICC)was used to evaluate the consistency of feature extraction.The Artificial Intelligence Kit(AK)post-processing software developed by GE Company was used to screen out the Radiomics features on T1 WI,T2WI and T1 WI enhancement pictures.There are totally 2688(896×3)Radiomics features in 7 categories,including histogram features,morphological features and texture features.Independent-Sample T test(normal distribution and variance)or Mann-Whitney U test(skewness distribution not neat or variance)were used to determine whether there were differences between the two groups of features.Univariate Logistic regression analysis was used to determine the relationship between different features and efficacy.Multi-factor Logistic regression analysis based on stepwise backward algorithm screening was used to determine whether features were independent predictors of efficacy.Radiomics models of T1 WI,T2WI,T1 WI enhancement and combined sequences were constructed respectively based on the screened features.Receiver operating characteristic(ROC)curve and area under curve(AUC)were used to evaluate the prediction performance of the model.Since this study was a single data set and all data were used for modeling,the Bootstrap method was used for 100 times resampling to conduct internal validation of the model,and the average accuracy,sensitivity and specificity of 100 times repeated validation were calculated.Results: After the above univariate and multivariate Logistic regression analysis,6features were selected from T1 WI as independent predictors of short-term efficacy,and the radiomics model was constructed to predict the efficacy of radiotherapy and chemotherapy,the accuracy,sensitivity,specificity and AUC were 0.7619,0.7705,0.7391 and 0.83 respectively.5 features were screened from T2 WI,the accuracy,sensitivity,specificity and AUC of the radiomics model were 0.7976,0.9180,0.4783 and0.77.8 features were screened from T1 WI enhancement,the accuracy,sensitivity,specificity and AUC of the radiomics model were 0.6786,0.5738,0.9565 and 0.82.13 features were screened from combined sequences,the accuracy,sensitivity,specificity and AUC of the radiomics model were 0.9167,0.9180,0.9130 and 0.95 respectively.The average accuracy,sensitivity and specificity of the validation set models validated within the group were 0.8805,0.9977 and 0.9985,respectively.Conclusion: The radiomics of MRI can predict the short-term efficacy of radiotherapy and chemotherapy for nasopharyngeal carcinoma,and the combined sequence-based model has the highest efficacy and can guide the individualized precise treatment.PART 2 A preliminary study of MRI radiomics in predicting metastasis and recurrence of nasopharyngeal carcinoma after radiotherapy and chemotherapyObjective: To explore whether MRI radiomics can predict the recurrence or metastasis of nasopharyngeal carcinoma after radiotherapy and chemotherapy.Methods: Retrospective analysis was performed on 52 patients with nasopharyngeal carcinoma first diagnosed by pathology in our hospital from January 2013 to January2018,with a follow-up period of 36 months.According to the prognosis,the patients were divided into:(1)relapse and/or metastasis group,24 cases;(2)without recurrence or metastasis group,28 cases.Fifty-two patients were randomly divided into a training set(40 cases,21 cases without recurrence or metastasis,19 cases with recurrence and/or metastasis)and a validation set(12 cases,7 cases without recurrence or metastasis,and 5cases with recurrence and/or metastasis).Two radiologists used ITK-Snap(version 3.6)to manually sketch the ROI layer by layer of the lesion area on the DICOM format images of T2 WI and T1 WI enhancement before treatment to obtain the 3D volume of interest(VOR)of the lesion.ICC was used to evaluate the consistency of feature extraction.The AK post-processing software developed by GE was used to screen out the radiomics features on the T2 WI and T1 WI enhancement.There are 2632(1316×2)features in 10 categories,including morphological features,first-order features,texture features and high-order features.The minimum redundancy and maximum correlation(m RMR)and minimum absolute contraction and selection operator(LASSO)regression were used to reduce the dimension to screen out the best radiomics features.The artificial neural network(ANN)was used to build the radiomics model of T2 WI and T1 WI enhancement respectively form the selected features,and the performance of the prediction model was evaluated by ROC curve and AUC.Results: Through the above m RMR and LASSO regression and ANN,6 features were screened out from T2 WI,and the training set AUC value,accuracy,sensitivity and specificity of the constructed image omics model were 0.778,0.757,0.789,and 0.722,respectively.The verification set AUC value,accuracy,sensitivity,and specificity were0.749,0.667,0.800,and 0.600.5 features were screened out in T1 WI enhancement.The AUC value of the training set of the radiomics model was 0.896,accuracy was 0.829,sensitivity was 0.833,and specificity was 0.824.The AUC value of the validation set was0.801,accuracy was 0.706,sensitivity was 0.833,and specificity was 0.636.Conclusion: MRI radiomics can predict the recurrence or metastasis of nasopharyngeal carcinoma,and the T1 WI enhancement Radiomics model has better predictive efficiency than the T2 WI Radiomics model.