| Part I Radiomic signature as a predictive factor for response to concurrent chemotherapy and radiationtherapy in advanced cervical cancerPurpose:To develop a radiomic model based on multiparametric MRI features in predicting the treatment response（PR/nonPR）before concurrent chemotherapy and radiationtherapy（CCRT）for advanced cervical cancer.Experimental Design:A total of 121 patients with advanced cervical carcinoma who underwent CCRT between December 2014 and June 2017 were enrolled in our study.All underwent 9 sequences MRI scan at baseline and then end of 4thweek during treatment.A total of 121 consecutive patients met the criteria.These patients were divided into a primary cohort（n=60）and a validation cohort（n=61）by using MR examination time point December 2015 as a cutoff.Two radiologists with 15 and 10years’experience in gynecological MR imaging independently assessed treatment response.According to Response Evaluation Criteria in Solid Tumors（RECIST）,the response results were divided into two groups:1)respond group including complete response（CR）and partial response（PR）;2)non-respond group including progressive disease（PD）and stable disease（SD）.We used all sequences original images archived in the Picture Archiving and Communication System（PACS）（Neusoft,V5.5.5.70228）We usedITK-SNAPsoftware（opensourcesoftware;www.itk-snap.org）for three-dimensional manual segmentation of the tumor habitat.All manual segmentations of the habitat were performed by a radiologist who had 10 years of experience,and each segmentation was validated by a senior radiologist,who had 15years of experience.We also randomly chose 25 images,which were segmented twice by the experienced radiologist with 10,15 years of experience for feature stability analysis.The radiomic features used in our study contained 4365 three-dimensional descriptors,485 for each habitat,including tumor intensity,shape and size,texture,and wavelet characteristics.The intraclass correlation coefficient（ICC）was calculated to analyze stability of the radiomic features.The Boruta package in R was used to assess the importance of each individual feature.By running the Random Forest algorithm iteratively and the“accepted”features are remained.We got 9 habitat signature from 9 sequences.Then the final primary cohort and validation cohort based on the 9 habitat signature were built.Finally,MRMR and support vector machine methods were performed to rank signatures and build a radiomic model.Statistical methods were based on R software（version 3.3.1）.hese prediction measures were computed using the R package pROC and MRMR features selection was applied using MRMRe package.The quantitative discrimination performance of the habitat signature was assessed using area under receiver operator characteristics curve（AUC）,classification accuracy,positive predict value（PPV）and negative predict value（NPV）in the primary cohort and validation cohort.The point on the ROC curve in the primary cohort for which the positive likelihood ratio was maximal was seen as the optimal cutoff threshold value and was also used in the validation cohort.The clinical risk factors with potential prognostic outcomes were identified in univariate analysis on the basis of the t-test for continuous variables and the chi-square test for categorical variables.P values<0.05 were considered to be statistical significance.Results:1.The clinical risk factors of patients were no significant differences between the respond group and the non-respond group in the primary and validation cohorts in terms of all clinical risk factors（P>0.05）by the univariate analysis.2.There was a significant association between the radiomic signature and the response of treatment in the primary（P<0.001）,which was confirmed in the validation cohort（P<0.001）.The result identified that only the radiomic signature was the independent predictor.3.The radiomic model,which consisted of 4 habitat signatures from Sagittal T2,Axial T2-FS,Sagittal T1 enhanced-MRI and Coronal T1 enhanced-MRI images,presented a good predictive performance,with an area under the curve of 0.896（95%CI,0.813–0.979）in the primary cohort and 0.853（95%CI,0.762–0.945）in the validation cohort.The accuracy were 0.850、0.754 in the primary and validation cohorts,respectively.The PPV,NPV were 0.933、0.767 in the primary cohort and 0.733、0.813 in the validation cohort.Conclusions:1.The clinical risk factors can not be used as predictors of response to CCRT.2.The radiomic signature was the only independent predictor of response to CCRT.3.Radiomic model employing features from multiple tumor habitats held the potential in predicting treatment response in patients with advanced cervical cancers before CCRT.These results illustrate a new discipline for improving medical decision-making and therapeutic strategies.Part II Radiomic signature as a predictive factor for lymph node metastasis in early-stage cervical cancerPurpose: Lymph node metastasis（LNM）is the most crucial independent risk factor for recurrence and death in patients with early-stage cervical cancer.The goal of the present study was to develop a radiomic signature of LN involvement based on sagittal T1contrast-enhanced（CE）and T2 MRI sequences.Experimental Design: We enrolled 143 pathologically confirmed early-stage cervical cancer.All underwent sagittal T1 CE and T2 MRI scans before treatment.Patients were randomly divided into 2 cohorts.One hundred patients were allocated to a primary cohort,while 43 patients were allocated to an independent validation cohort.We used sagittal T1 enhanced and T2 MRI DICOM original images archived in the PACS.The segmentation of a region of interest（ROI）is essential for the extraction of quantitative radiomic features.A radiologist with 10 years of experience used the ITK-SNAP open-source software for 3D manual segmentation.A senior radiologist with 15 years of experience validated all segmentations.A total of 970 radiomic features and 7 clinical characteristics were extracted from each patient,485 for each sequence.The features were divided into 4 groups:（I）tumor intensity,（II）shape and size,（III）texture,and（IV）wavelet characteristics.MRMR and SVM were applied to select features and construct a radiomic signature.The performance of the radiomic signature was assessed in primary and validation cohorts.Statistical methods were based on the R analysis platform（version 3.3.1）.In univariate analysis,the Mann-Whitney U test and the chi-square test were used for continuous and categorical variables,respectively,to test the performance of clinical characteristics and potential prognostic outcomes.P values<0.05 were considered to indicate statistical significance.The area under the receiver operating characteristics curve（ROC AUC）,classification accuracy,sensitivity,and specificity were used to assess the quantitative discrimination performance.The point on the ROC curve with the maximum positive likelihood ratio was considered the optimal cutoff threshold value.These prediction measures were computed using the R package p ROC,and MRMR feature selection was done using the MRMRe package.Results:1.The characteristics of all patients,such as age,pregnancy number,parturition number are no significant differences in any characteristics（P > 0.05）between the LNM and non-LNM groups in both the primary and validation cohorts.2.The radiomic signature,consisting of 10 features,showed a good discrimination between LNM and non-LNM groups.The area under the receiver operating characteristic curve was 0.753（95% CI,0.656–0.850）in the primary cohort and 0.754（95% CI,0584–0.924）in the validation cohort.The accuracy,PPV,NPV were 0.753、0.750、0.750 in the primary cohort and 0.721、0.714、0.724 in the validation cohort,respectively.Conclusions:1.The clinical risk factors can not be used as predictors for lymph node metastasis in early-stage cervical cancer.2.A multiple-sequence MRI radiomic signature can be used as a noninvasive biomarker for preoperative assessment of LN status and potentially influence the therapeutic decision making in early-stage cervical cancer patients. |
PDF Full Text Request