Prediction Of Prognosis Of Primary Nasopharyngeal Carcinoma Based On Multiparametric MRI And Radiomic Analysis

The first part Prediction of Local Recurrence Risk of Nasopharyngeal Carcinoma based on Multiparametric MRIPurpose: Radiotherapy is the standard treatment for nasopharyngeal carcinoma.At present,intensity modulated radiation therapy(IMRT)is the main treatment.Although the improvement of treatment methods has significantly improved the level of disease control,local residual and local recurrence are not uncommon after treatment of nasopharyngeal carcinoma,which affects the prognosis of patients.MRI is the first choice for the diagnosis,staging,evaluation of treatment response and efficacy monitoring of nasopharyngeal carcinoma.Previous studies have reported that MRI is helpful to diagnose and evaluate local residual and local recurrent diseases of nasopharyngeal carcinoma after treatment,but there is no clear consensus on the relationship between them.Therefore,based on the conventional MRI findings of nasopharyngeal carcinoma,this study investigated the relationship between local residual and local recurrence after treatment,and established a model to predict the local recurrence of nasopharyngeal carcinoma,so as to assist the clinical early identification of patients with recurrence risk,so as to benefit from early treatment and improve the prognosis of patients.Materials and methods:1.Clinical data135 patients with primary nasopharyngeal carcinoma confirmed by nasopharyngeal biopsy from April 2013 to April 2018 were collected retrospectively.The clinical data of patients,including age,gender,KPS score,T stage of nasopharyngeal carcinoma,treatment mode and pathological type,were obtained from clinical electronic medical records.2.MRI scanningThe Siemens magnetem Verio 3.0T system magnetic resonance scanner was used for images acquisition.MRI scan included plain scan and enhancement scan,in which plain scan included axial T1 WI,axial T2 WI,sagittal T1 WI and coronal T2 WI,and enhancement scan included axial,coronal and sagittal T1 WI.3.MRI image evaluation and feature recording The MRI data of the first diagnosis and 3-6 months after treatment were collected and read jointly by a head and neck radiologist with more than 15 years of work experience and two imaging physicians with more than 8 years of work experience.The features evaluated included nasopharyngeal(NC)lesions,nasal oral(NOC)involvement,parapharyngeal space(PS)involvement,skull base bone erosion(SBBE),masticatory muscle space invasion(MSI),MRI detected cranial nerve invasion(MDCNI),paranasal sinus invasion(PSI),and retropharyngeal lymph node(RLN)involvement.According to the UICC/AJCC staging 8th Edition/China staging2017 edition standard of nasopharyngeal carcinoma,the T stage of nasopharyngeal tumors was reevaluated by MRI 3-6 months after radiotherapy and recorded as r T stage.4.Follow up and clinical end points The local recurrence of patients was recorded according to regular follow up.The clinical endpoint of this study was local recurrence-free survival(LRFS),which was defined as the time from the beginning of treatment to the observation of local recurrence.This study confirmed that the follow-up time of LRFS was at least 24 months.Patients with local recurrence needed to be followed up by endoscopic biopsy or enhanced MRI of skull base nasopharynx.5.Statistical analysis Categorical variables in MRI and clinical features were tested by c2 test or Fisher exact test,and continuous variables were tested by independent sample t-test.Logistic regression analysis was used to screen the independent predictors of local recurrence of nasopharyngeal carcinoma.The predictive efficacy of the model was analyzed by receiver operating characteristic(ROC)curve.Kaplan-Meier survival analysis was used to analyze the relationship between variables and LRFS.All statistical tests were bilateral,and P<0.05 was the difference with statistical significance.Results:1.r T stage and NC residual lesions were independent risk factors for local recurrence of nasopharyngeal carcinoma.2.Compared with individuals with r T stage at r Tx and r T0 stage,r T3-r T4 stage increased the risk of local recurrence in patients with nasopharyngeal carcinoma(OR=5.752,95%CI:1.085-30.483).3.Compared with individuals without residual lesions in NC,the presence of residual lesions in NC increased the risk of local recurrence of nasopharyngeal carcinoma(OR=11.402,95%CI:3.520-36.931).4.The prediction model based on r T stage and NC residual lesions was Logit(P)=-2.049+1.750*(r T3-r T4)-0.364*(r T1-r T2)+2.434*NC.ROC curve showed that the prediction model jointly established by the two had the largest AUC,with a value of0.834(95%CI:0.753-0.916).Conclusion:1.r T stage and NC residual lesions were independent risk factors for local recurrence of nasopharyngeal carcinoma.2.The prediction model based on r T stage and NC residual lesions had good efficiency for predicting local recurrence of nasopharyngeal carcinoma.It could provide clinicians with individualized prediction probability of local recurrence and promote the improvement of follow-up and treatment of patients with high recurrence risk.The second part Prediction of Risk of Disease Progression of Nasopharyngeal Carcinoma based on Multiparametric MRI and Radiomic AnalysisObjective: Local recurrence and distant metastasis are the main causes of treatment failure of nasopharyngeal carcinoma and reduce the overall survival.Therefore,it is urgent for clinical practice to accurately predict the adverse prognosis of patients with nasopharyngeal carcinoma.Radiomics is a new and rapidly developing hot field of computer technology.It can comprehensively and quantitatively evaluate the heterogeneity of tumors and predict the biological characteristics of tumors.Based on this,this study discussed the feasibility and clinical application value of rad-score based on multiparametric MRI to predict the risk of early disease progression of nasopharyngeal carcinoma,and established a prediction model combined with the characteristics of conventional MRI,so as to provide an important basis for clinical individualized treatment decision-making.Materials and methods:1.Clinical data155 patients with primary nasopharyngeal carcinoma admitted to our hospital from April 2013 to September 2018 were retrospectively collected.The TNM stage was stage I-IV.The clinical data of the first visit were obtained from the clinical records,including gender,age,histological type,T stage,N stage,total clinical stage and treatment mode.All patients received a complete course of radiotherapy after admission.2.MRI scanning All patients used the Siemens magnetem Verio 3.0T system magnetic resonance scanner for plain scan and enhanced scanning imaging,and the scanning sequence acquisition was the same as the first part.3.Evaluation of MRI features Conventional follow-up MRI 3-6 months after treatment was compared with pretreatment MRI to evaluate the residual status of nasopharyngeal carcinoma lesions.It was recorded as tumor residual and no tumor residual.4.Follow up and clinical end points Local recurrence,metastasis and death were recorded according to regular follow-up.This study took progression free survival(PFS)as the clinical endpoint,and confirmed that the follow-up time of PFS was at least 36 months.PFS was defined as the time from the first day of treatment to disease progression(including recurrence and distant metastasis)and death from any cause,or the last follow-up(called censored data).All local recurrence cases were diagnosed according to the results of nasopharyngeal biopsy or nasopharyngeal+skull MRI.The diagnosis of neck recurrence depended on clinical neck physical examination combined with puncture pathology or neck MRI.The diagnosis of distant metastasis was based on clinical and imaging examination,such as chest X-ray,whole-body bone scan,MRI,CT,PET/CT and abdominal ultrasound.5.Image segmentation and extraction and selection of radiomics features The DICOM format images of axial T1 WI,T2WI and T1WI+C sequences of patients before treatment were retrieved and exported from the hospital PACS to the free ITKsnap software.The 3D region of interest(ROI)of the whole tumor was manually outlined by two doctors.The pyradiomics package in Python 3.6.5 was used to extract the radiomics features from the segmented nasopharyngeal carcinoma lesions,and the same number of features were extracted from each sequence.For the extracted radiomics features,ICC was first calculated to evaluate the consistency of feature extraction among observers,and the features with ICC greater than 0.75 were selected for the next analysis.The patients were randomly divided into training cohort(108cases)and validation cohort(47 cases)according to the ratio of 7:3.Using the data of the training cohort,firstly,Mann Whitney U test was performed on the radiomics features included in the analysis to identify the difference between the disease progression group and the non-progression group,and the features with P<0.05 were selected.Then the least absolute shrinkage and selection operator algorithm(LASSO)was used to screen the features,and then a logistic regression model with Akaike Information Criterion(AIC)was further used to select the features finally.6.Construction,verification and prognostic value of rad-score The rad-score was obtained by linear combination of the radiomics features screened by LASSO and its corresponding coefficients.In the training cohort,based on the features screened by LASSO algorithm,the rad-scores of T1 WI,T2WI,T1WI+C and multi-sequence fusion were established by logistic regression,and tested in the validation cohort.Area under curve(AUC),accuracy,sensitivity and specificity were used for quantitative comparison.The potential correlation between PFS and radscore was first evaluated in the training cohort and then confirmed in the validation cohort.Patients were divided into high and low risk groups based on the median radscore,and Kaplan-Meier survival analysis was performed in both groups.7.Construction,validation and prognostic value of nomogram The clinical features and conventional MRI features were compared between the disease progression and non-progression groups,and the statistically different features and rad-score were included in the multivariate logistic regression analysis.The independent predictors were screened,combined with the independent predictors to construct the nomogram model,and tested in the validation cohort.Draw ROC curve,calibration curve and decision curve to evaluate nomogram model.Patients were divided into high and low risk groups based on the median score of nomogram model,and Kaplan-Meier survival analysis was performed in the two groups.8.Statistical method SPSS software(version 24.0)and R language "RMS" software package(version 3.5.1)were used for statistical analysis.Independent sample t-test or Mann-Whitney U test were used for measurement data.Counting data used c 2 test or Fisher exact test.Univariate analysis of clinical and MRI features with differences between the progression and non-progression groups and rad-score were included in the multivariate logistic regression analysis to screen independent predictors and construct nomogram.Kaplan-Meier method was used for survival analysis.All statistical tests were bilateral,and P<0.05 was the difference with statistical significance.Results:1.Univariate analysis showed that the total clinical stage and MRI-residual were statistically different between the disease progression group and the non-progression group.2.Compared with a single sequence,multi-sequence fusion rad-score was the best in predicting the disease progression of nasopharyngeal carcinoma.The AUC of the training cohort was 0.890(95%CI:0.828-0.952),The AUC of the validation cohort was 0.808(95%CI:0.684-0.935).3.Multivariate logistic regression analysis showed that MRI-residual and rad-score were independent predictors.The nomogram model constructed by the two groups was the best in predicting the disease progression of nasopharyngeal carcinoma.The AUC of the training cohort was 0.923(95%CI:0.874-0.973),The AUC of the validation cohort was 0.882(95%CI:0.596-0.844).The calibration curve showed that the observed and predicted values of nomogram were in good agreement.The decision curve showed that the model had clinical net benefit when the threshold probability was 0.19-0.95.Conclusion:1.Multi-sequence fusion rad-score and MRI-residual were independent predictors of disease progression of nasopharyngeal carcinoma.Multi-sequence fusion rad-score was effective in predicting the progression of nasopharyngeal carcinoma.2.The nomogram model based on rad-score combined with MRI-residual further improved the prediction efficiency.It was expected to assist the decision-making of clinical individualized diagnosis and treatment,make patients get the maximum treatment benefit and improve the quality of life and survival.