Analysis Of Prognosis And Efficacy Based On Magnetic Resonance Radiomics In The Nasopharyngeal Carcinoma

Part I Application value of magnetic resonance multi-parameter radiomics and clinical histograms in evaluating the sensitivity of neoadjuvant chemotherapy in nasopharyngeal carcinoma Objective: To predict the sensitivity of nasopharyngeal carcinoma(NPC)to neoadjuvant chemotherapy(NACT)based on magnetic resonance(MR)radiomics and clinical nomograms prior to NACT.Materials and Methods: From January 2014 to July 2015,284 consecutive patients with pathologically confirmed NPC underwent 3.0 T MR imaging(MRI)before initiating NACT.The patients’ data were randomly assigned to a training set(n=200)or a test set(n=84)at a ratio of 7:3.The clinical data included sex,tumor(T)stage,lymph node(N)stage,American Joint Committee on Cancer(AJCC)stage,and the plasma concentration of Epstein-Barr virus(EBV)DNA.The regions of interest(ROI)were manually segmented on the axial T2-weighted imaging(T2WI),Diffusion Weighted Imaging(DWI)and enhanced T1-weighted imaging(T1WI)sequences using ITK-SNAP software.The radiomics data were post-processed using AK software.Moreover,the Maximum Relevance Minimum Redundancy(m RMR)algorithm and the Least Absolute Shrinkage and Selection Operator(LASSO)were adopted for dimensionality reduction to screen for the features that best predicted the treatment efficacy,and clinical risk factors were used in combination with radiomics scores(Rad-scores)to construct the clinical radiomics-based nomogram.De Long’s test was utilized to compare the area under the curve(AUC)values of the clinical radiomics-based nomogram,radiomics model,and clinical nomogram.Decision curve analysis(DCA)was employed to evaluate each model’s net benefit.Results: The clinical nomogram was constructed based on data from patients who were randomly assigned according to T2 WI,DWI and enhanced T1 WI sequences.In the training set,the T2 WI sequence-based clinical radiomics nomogram and the radiomics model outperformed the clinical nomogram in predicting the NACT efficacy(AUC,0.81 vs 0.60,P=0.001;and 0.76 vs 0.60,P=0.030).These findings were well-verified in the test set.The DWI sequence-based clinical radiomics nomogram outperformed the clinical histogram in predicting the NACT efficacy(AUC,0.78 vs 0.68;P= 0.010).The enhanced T1 WI sequence-based clinical radiomics nomogram exhibited better performance in predicting treatment efficacy than the clinical nomogram(AUC,0.79 vs0.62,respectively;P=0.0001).The DCA revealed that,the T2WI-,DWI-and clinical radiomics-based histograms gained net benefit in predicting the NACT efficacy.Conclusion: The clinical radiomics-based nomogram improved the prediction of NACT efficacy,with the T2 WI sequence-based clinical radiomics achieving the best effect.Part II Value of magnetic resonance multi-parameter radiomics and clinical histograms in predicting of prognosis in nasopharyngeal carcinomaObjective:To explore the value of magnetic resonance multi-sequence radiomics and clinical histograms before treatment in predicting the long-term recurrence of nasopharyngeal carcinoma(NPC).Materials and methods:This study collected 284 newly diagnosed NPC patients,including 186 with no recurrence within 5 years and 98 with recurrence(45 with metastasis and 54 with local recurrence).The remaining materials and methods were the same as those in Part one.Results:The clinical histogram was constructed based on patients randomly assigned according to T2 WI,DWI and T1 WI enhanced sequences.In the training set,the T2WI-based clinical radiomics histogram and radiomics model had superior efficacy in predicting NPC recurrence to the clinical histogram(AUC,0.81 vs 0.60;P<0.0001)(AUC.In the DWI sequence-based training set,the clinical radiomics histogram was superior to the clinical histogram(AUC,0.74 vs 0.63;P= 0.004).The above findings were well verified in the test set.The clinical radiomics histogram model constructed based on the T1 WI enhanced sequence deviated from the fitting(Hosmer-Lemeshow test,training set P= 0.024,test set P = 0.031).There was no statistical significance between the radiomics histogram and clinical histogram model in the training set(AUC,0.74 vs 0.70;P= 0.543).DCA revealed that the T2 WI sequence-based radiomics and clinical radiomics model and the DWI-based clinical radiomics histogram gained net benefit in the prediction of long-term recurrence.Conclusion:T2WI-based clinical radiomics histogram and radiomics model effectively predicted local recurrence and distant metastasis,while the T1 WI enhanced sequence-based model deviated from the fitting.Part III Prediction of prognosis of nasopharyngeal carcinoma after neoadjuvant chemotherapy based on MR radiomics and multimodal machine learningObjective : To explore the value of magnetic resonance radiomics and multimodal machine learning classifiers post-neoadjuvant chemotherapy(NACT)in predicting the long-term recurrence of nasopharyngeal carcinoma(NPC).Materials and methods:Altogether 271 consecutive NPC patients after NACT were selected,including 183 without recurrence within 5 years and 88 with recurrence.All patients were randomly divided into training set(n=185)and test set(n=86).The regions of interest(ROI)were delineated on the axial T2 WI and T1 WI enhanced sequences,and LASSO was employed for dimensionality reduction to construct the final model.In addition,seven machine learning classifiers,including KNN,Ada Boost,RF,SVM,MLR,NB and LDA,were analyzed.The area under the curve(AUC)values were used to evaluate the efficacy to predict 5-year recurrence.Results:The AUC values of diverse T2 WI sequence-based machine learning classifiers in predicting NPC recurrence ranged from 0.87 to 1,among which,RF achieved AUC=1,with the accuracy,specificity and sensitivity reaching 100%.KNN had the second best performance,with the AUC=0.92,accurate of 93%,sensitivity of 89%,specificity of 95%.Classifier RF had the highest prediction performance.The AUC values of diverse machine learning classifiers in test set ranged from 0.79 to 0.88.On the T1WI+C sequence,the AUC values of diverse machine learning classifiers were between 0.85 and 1.The AUC was 1 for RF,while the accuracy,specificity and sensitivity all reached 100%,and the AUC values of diverse machine learning methods in test set were between 0.81 and 0.88.Conclusion:After NACT,the T2 WI and T1 WI enhanced sequences-based radiomics and multimodal machine learning accurately predicted the long-term(5-year)recurrence.