Prediction Of Efficacy And Prognosis Of Concurrent Chemoradiotherapy For Locally Advanced Cervical Squamous Cell Carcinoma Based On DCE-MRI And Radiomics Analysis

The first part DCE-MRI in locally advanced cervical squamous cell carcinoma:correlation between pharmacokinetic parameters and prognosis of concurrent chemoradiotherapyObjective:Cervical squamous cell carcinoma is the most common pathological type of cervical cancer,and concurrent chemoradiotherapy has become the standard treatment of locally advanced cervical squamous cell carcinoma.If the prognosis can be predicted before the treatment,it can guide the clinical strategy.Dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)describes the pharmacokinetic process of contrast agent in tissues and can generate multiple pharmacokinetic parameters for evaluating the tissue microenvironment of cervical squamous cell carcinoma quantitatively.Currently,only a few studies have addressed DCE-MRI pharmacokinetic parameters for the prognosis prediction of cervical cancer after concurrent chemoradiotherapy,and the clinicopathological features are not considered,while the prognosis prediction of cervical squamous cell carcinoma has not been confirmed.Thus,the present study aimed to investigate the predictive value of DCE-MRI pharmacokinetic parameters on the prognosis of concurrent chemoradiotherapy for locally advanced cervical squamous cell carcinoma.Methods:1.Clinical dataA total of 131 patients with cervical squamous cell carcinoma(FIGO stage ?B-?A stage),who received concurrent chemoradiotherapy from January 1,2013 to March 1,2014,underwent DCE-MRI within 2 weeks before the treatment.2.MR imagingAll MRI scans were conducted on a Siemens 3.0T MRI scanner.The acquisition parameter ranges were sagittal TSE T2WI,axial TFS T2WI and axial T1WI.DCE-MRI scans were performed using a continuous contrast-enhanced multiphase acquisition with a total of 30 phases.3.Concurrent chemoradiotherapyRadiotherapy was administered by linear accelerator 6MV X-ray,including three-dimensional conformal intensity-modulated radiotherapy and intracavity after-loading treatment.Chemotherapy was administered by paclitaxel+cisplatin(TP regimen),paclitaxel+carboplatin(TC regimen),and docetaxel+cisplatin(DP regimen).4.Evaluation of tumor treatment responseAccording to the response evaluation criteria in solid tumors(RECIST 1.1),the evaluation included complete response(CR),partial response(PR),progressive disease(PD)and stable disease(SD).5.Image analysis and post-processingAll datas were processed using Tissue 4D software.Two radiologists drew the region of interest(ROI)along the edge of lesion with the maximum diameter on the pseudo-color maps of pharmacokinetic parameters.Tofts model and arterial input function(AIF)in tissue 4D were used to calculate the K^trans,Kep,and Ve.The ROI was averaged by outlining three times.6.Follow-UpThe local recurrence,metastasis,and death of patients were recorded by regular follow-up,and the disease-free survival(DFS)was calculated.The follow-up deadline was June 1,2019.According to the follow-up results,the enrolled patients were divided into non-recurrence and recurrence groups.7.Statistical analysisThe intraclass correlation coefficient(ICC)was calculated for the two sets of results.?2 test was used for the correlation test of categorical variables.Mann-Whitney U test was used to compare the difference between non-recurrence group and recurrence group after concurrent chemoradiotherapy.Receiver operating characteristic(ROC)curve was drawn to analyze the diagnostic efficacy of pharmacokinetic parameters for selecting the optimal parameter and its diagnostic threshold.Pearson's correlation analysis was used to analyze the correlation between pharmacokinetic parameters.Kaplan-Meier method and Cox multivariate regression analysis were used to analyze the relationship between the variables and DFS.Results:1.Baseline data and single-factor analysis FIGO stage,tumor differentiation,tumor diameter,response to concurrent chemoradiotherapy and pharmacokinetic parameters(K^transand Ve)were the influencing factors for DFS of locally advanced cervical squamous cell carcinoma with concurrent chemoradiotherapy(P<0.05).2.Reproducibility results of DCE-MRI pharmacokinetic parameters The ICCs of K^trans,Kep,and Ve were 0.899,0.802,and 0.815,respectively.3.Efficacy of DCE-MRI pharmacokinetic parameters in predicting recurrence and metastasis of locally advanced cervical squamous cell carcinoma K^trans and Ve of non-recurrence group were 0.692±0.177 min-1 and 1.029±0.393,significantly higher than those of recurrence group with 0.457±0.107 min-1 and 0.725±0.112(P<0.05).However,no statistically significant difference was observed in Kep between non-recurrence group and recurrence group(P>0.05).AUCs of K^trans,Kep and Ve were 0.869,0.618 and 0.771,with good diagnostic efficacy.4.Cox multivariate analysis for DFS K^trans was an independent factor for DFS of locally advanced cervical squamous cell carcinoma(HR=2.565,P<0.05).The smaller the K^trans value was,the greater the probability of recurrence and metastasis was.Conclusion:1.DCE-MRI pharmacokinetic parameters have good stability.2.K^trans is an independent factor for DFS of patients with locally advanced cervical squamous cell carcinoma who underwent concurrent chemoradiotherapy.This could provide accurate guidelines and scientific basis for the prognosis evaluation and treatment of locally advanced cervical squamous cell carcinoma.The second part Evaluation of multi-sequence MRI radiomics in predicting the response to concurrent chemoradiotherapy for locally advanced cervical squamous cell carcinomaObjective:Heterogeneity within locally advanced cervical squamous cell carcinoma may be a key factor affecting the response to concurrent chemoradiotherapy.Accurate prediction of the response before treatment may provide a basis for individualized concurrent chemoradiotherapy,leading to more effective treatment for locally advanced cervical squamous cell carcinoma with a more hostile tumor microenvironment.Radiomics can quantitatively analyze the heterogeneity of locally advanced cervical squamous cell carcinoma by parameters extracted from imaging images in a non-invasive manner,and has great potential for application in predicting the response to concurrent chemoradiotherapy.At present,the results of MRI radiomics in predicting the response to chemoradiotherapy for cervical cancer are not consistent,and there is no research on predicting the response to concurrent chemoradiotherapy for locally advanced cervical squamous cell carcinoma based on multi-sequence radiomics model.In this study,we established a multi-sequence radiomics model,aiming to investigate the predictive value of MRI radiomics model in assessing the response to concurrent chemoradiotherapy for locally advanced cervical squamous cell carcinoma.Methods:1.Clinical dataA total of 367 patients with cervical squamous cell carcinoma(FIGO stage ?B-?A stage)confirmed by pathology from January 1,2013 to June 30,2019 were retrospectively collected.They were unable to undergo surgery and received complete concurrent chemoradiotherapy.Pelvic plain MRI,DWI and DCE-MRI were performed within 2 weeks before treatment.2.MR imagingAll MRI scans were conducted on Siemens 3.0T MRI scanners.Conventional MRI and DCE-MRI were performed as in the first part.B values of DWI were Os/mm2 and 800s/mm2,respectively.3.Concurrent chemoradiotherapyThe same with the first part.4.Evaluation of tumor treatment responseTwo radiologists combined conventional MRI,DWI,and DCE-MRI images to perform a comprehensive analysis of MRI images within 2 weeks before concurrent chemoradiotherapy and at the end of the 4th week of treatment,and evaluated them according to RECIST 1.1 criteria.Patients were divided into CR group and non-CR group(including PR and SD).No PD cases were in this study.5.Image processing and feature extractionDWI,T2WI and enhanced T1WI delayed phase images were used for radiomics feature extraction.A radiologist combined the three sequence images and drew ROI layer by layer semi-automatically or manually,and this was then checked by another radiologist to form the VOI.1906 radiomics features based on the original and pre-processed images were extracted from DWI,T2WI and enhanced T1WI,respectively,and normalized by Z-score.6.Feature selection and radiomics prediction model establishmentFeature correlation analysis and tree model were used for feature selection.The patients were divided into a training set(n=256)and a validation set(n=111)via a randomized split modeling validation method at a ratio of 7:3.Machine learning was performed using three classifier learning algorithms,namely Logistic regression(LR),Support vector machine(SVM)and Random forest(RF).The diagnostic efficacy of the three classifiers was evaluated by comparing AUCs.The DWI,T2WI and enhanced T1WI radiomics features with higher diagnostic efficiency and less correlation were selected respectively to build a multi-sequence combined model,and the diagnostic efficacy of the three single sequence models and the multi-sequence combined model was evaluated by comparing AUCs.The clinical application value of the multi-sequence combined model was evaluated by decision analysis curve(DCA).7.Statistical analysisThe Dr.Wise multimodal research platform was applied for statistical analysis.Independent sample t-test was used for conformity to normal distribution,and Wilcoxon test was used for non-conformity to normal distribution.The chi-square test was used for comparison between two groups of categorical variables.The differences in AUCs for machine learning of the three classifiers were compared by DeLong test,as well as comparing the differences in AUCs of the three single models and multi-sequence combined sequence model.Result:1.Baseline dataOf the 367 patients with locally advanced cervical squamous cell carcinoma,247 cases were in CR group and 120 cases were in non-CR group.There were 172 cases in CR group and 84 cases in non-CR group in the training set(n=256),and 75 cases in CR group and 36 cases in non-CR group in the validation set(n=111).There was no significant difference in age,FIGO stage,tumor differentiation and tumor diameter(P>0.05).2.Performance evaluation of different classifier modelsThe LR model had the best performance.Diagnostic efficacy of the SVM model was slightly lower than that of the LR model.The RF model was over-fitted.3.Performance evaluation of single sequence models and multi-sequence modelThe AUCs of DWI,T2WI and enhanced T1WI radiomics models had no statistical difference(P>0.05),while the AUC of the multi-sequence combined model in the training set was 0.86,higher than that of the three single sequence models with statistically differences(P<0.05).The DCA showed that within a certain threshold probability range,the multi-sequence combined model yielded a net benefit.Conclusion:Using radiomics features extracted from multi-sequence MR images,feature selecting and machine learning,and building radiomics models,have predictive value for assessing the response of concurrent chemoradiotherapy for locally advanced cervical squamous cell carcinoma.The multi-sequence combined model can improve the diagnostic efficacy and help guide clinical decision-making.