| Part ⅠBased on mpMRI lymph node image segmentation to extract Radiomics parameters,construct Radiomics labels to predict cervical cancer lymph node metastasisObjective:By analysis the Radiomics features of lymph node images extracted from multi-parameters magnetic resonance imaging(mpMRI)before treatment,construct Radiomics signature to realize non-invasive early prediction of cervical cancer lymph node metastasis.Methods:This study prospectively enrolled 189 patients with cervical cancer.All patients have finished pelvic 3.0 T MRI examination according to the standard protocol of cervical cancer.Surgery was performed within 10 days after the MRI scan.Using the image-surgery-pathological comparison method,a total of 162 lymph nodes corresponding to the image and pathology were obtained,of which 86 were positive lymph nodes confirmed by pathology(taken from 27 patients with lymph node metastasis)and 76 were negative lymph nodes(taken from 30 patients without Patients with lymph node metastasis).The 162 lymph nodes that meet the criteria for entry were separately performed by a physician with more than 10 years of diagnostic experience in our department using ITK-SNAP software to complete the 3D manual segmentation of coronary T2WI and transverse T1WI enhanced arterial phase andvenous phase sequence lymph node images,and then use The Anaconda 3 software pyRadiomics package extracts Radiomics from the above-mentioned VOI(according to the European and American Radiomics feature standard IBSI standard).Each sequence extracts 851 Radiomics features of lymph nodes.Firstly,the imaging omics tag Radscore was constructed based on a single MRI sequence,and then the three sequences were combined to construct the imaging omics tag Radscore.The imaging omics label construction method is to first randomly divide all lymph nodes into a training group and a verification group at 7:3,with 115 and 47 lymph nodes respectively,of which 61 are metastatic lymph nodes in the training group and 54 are non-metastatic;in the verification group There were 25 metastatic lymph nodes and 22 non-metastatic lymph nodes.Then,the maximum correlation and minimum redundancy mRMR is used to remove redundancy and impurity.After keeping 20 features,LASSO is used to perform feature dimensionality reduction,and the feature number corresponding to the minimum penalty coefficient logλ is selected to construct the Radiomics signature.Results:1.Based on T2WI segmentation of lymphatic structure to build Radscore,take the minimum penalty coefficient logλ=0.033 and finally corresponding 15 imaging features to construct the Radscore to identify whether lymph nodes are metastasized.AUC in the training set is 0.88(95%CI,0.81).-0.94),0.87(95%CI,0.78-0.97)in the validation set.The.accuracy,sensitivity,and specificity were 0.948,0.950;0.944 in the training group,and 0.936,0.893,and 1.000 in the verification group.2.Based on T1WI to enhance the arterial phase segmentation of lymphatic structure to build Radscore,take the smallest penalty coefficient logλ=0.012 and finally corresponding 13 Radiomics features to construct the Radscore to identify whether the lymph nodes are metastasized.AUC is 0.90(95%)in the training set.CI,0.85-0.95),0.89 in the validation set(95%Cl,0.79-0.98).Accuracy,sensitivity,and specificity were 0.947,0.936,and 0.961 in the training group,and 0.93WI6,0.923,and 0.952 in the syndrome group.3.Based on T1WI to enhance the venous phase segmentation lymphatic structure to build Radscore,take the minimum penalty coefficient logλ=0.009 and finally corresponding 15 Radiomics features to construct the Radscore to identify whether the lymph nodes are metastasized.AUC is 0.92(95%)in the training set.CI,0.88-0.97),0.91(95%CI,0.82-1.00)in the validation set.The accuracy,sensitivity,and specificity were 0.930,0.920,and 0.942 in the training group,and 0.978,0.887,and 1.000 in the syndrome group.4.Based on T2WI,T1WI-enhanced arterial phase,T1WI-enhanced venous phase segmentation of lymphatic structure to build Radscore,take the smallest penalty coefficient logλ=0.025 and finally the corresponding 13 Radiomics features are constructed to construct the Radscore,which is used to distinguish lymph node metastasis or not.Better performance,AUC is 0.94(95%CI,0.90-0.98)in the training set and 0.93(95%CI,0.86-1.00)in the validation set.The accuracy,sensitivity,and specificity were 0.947,0.923,and 0.980 in the training group,and 0.936,0.958,and 0.913 in the syndrome group.Conclusion:Based on pre-treatment T2WI,T1WI-enhanced arterial phase and venous phase images combined with lymph node image segmentation and extraction,the Rad-score can be used to early non-invasively predict cervical cancer lymph node metastasis,and can be used as a non-invasive,quantitative imaging marker to assist the cervix Preoperative staging of cancer.Part ⅡConstructing Radscore to predict lymph node metastasis based on the segmentation of cervical cancer mpMRI tumor image and the extraction of Radiomics parametersObjective:The pre-treatment Radiomic features extracted from mpMRI cervical tumors are characterized to reduce redundancy to construct the optimal Radscore combined with clinical parameter modeling to achieve non-invasive preoperative prediction of cervical cancer lymph node metastasis.Methods:This This study retrospectively enrolled 630 patients who were diagnosed with cervical cancer(stage Ⅰbl-Ⅱb)and underwent C-type Radical resection and pelvic lymph node resection in Yunnan Cancer Hospital from October 2012 to December 2017.All patients have finished pelvic MRI examination according to the standard protocol of cervical cancer.Patients who finally met the inclusion criteria were divided into surgery only group and neoadjuvant chemotherapy combined with surgery group according to different treatment regimens and the lymphatic metastasis of cancer cells of each group was investigated.Patients in different group were randomly divided into training group and validation group in a ratio of 7:3.Three dimensional manual segmentation of the cervical axial T2WI,transverse position T1WI enhanced arterial and venous phase sequence of cervical tumor Image were performed by physicians in my department who has 10 years experience in diagnosis by using the ITK-SNAP software,then according to the imaging markers Standardization Initiative(Image Biomarker Standardization Initiative,IBSI)USES the Anaconda 3 software package import pyRadiomics features extracted from the aforementioned VOI Image characteristics),For each sequence,851 Radiomics features of tumor lesions were extracted.First,the imaging omics label RAD score was constructed based on a single MRI sequence,and then the optimal diagnostic efficacy MRI sequence was selected to construct the Radiomics label Rad score.Then,the Nomogram visualization model was used to construct the multiple logistic regression model combined with clinical parameters to predict lymph nodes.The clinical parameters were screened by progressive multiple logistic.Results:1.The Rad score was constructed based on tumor image segmentation and radiomics parameters extracted from mpMRI to evaluate lymph node metastasis in cervical cancer patients recived neoadjuvant chemotherapy combined with surgery.1).The Rad score was constructed based on T2WI segmentation of cervical cancer to evaluate lymph node metastasis,the AUC was 0.751(95%CI,0.623-0.878)and 0.765(95%CI,0.526-1.000)in the training cohort and validation cohort respectively.The accuracy,sensitivity and specificity were 0.678,0.571 and 0.941 in the training cohort,and 0.782,0.823 and 0.667 in the validation cohort,respectively.2).The Rad score was constructed based on T1WI enhanced arterial phase segmentation of cervical tumors to assess lymph node metastasis.The AUC was 0.860(95%CI,0.7555-0.965)in the training cohort and 0.902(95%CI,0.7544-1.000)in the validation cohort.The accuracy,sensitivity and specificity were 0.864,0.905 and 0.764 in the training cohort,and 0.913,0.941 and 0.833 in the validation cohort,respectively.3).The Rad score was constructed based on T1WI enhanced venous segment of cervical tumors to evaluate lymph node metastasis The AUC was 0.833(95%CI,0.720--0.947)in the training cohort and 0.853(95%CI,0.693--1.000)in the validation cohort.The accuracy,sensitivity and specificity were 0.814,0.810 and 0.823 in the training group,and 0.826,0.764 and 0.600 in the validation group,respectively4).The multivariate logistic regression model Nomogram visualization was developed by using the image segmentation radiomics tag Rad score based on the optimal diagnostic performance T1WI enhanced arterial cervical tumor in combination with clinical parameters(FIGO staging and preoperative MRI measured tumor diameter line),which was used for the assessment of lymph node metastasis,The AUC in the training cohort was 0.90(95%CI,0.82 0.98),and the AUC in the validation cohort was 0.91(95%Cl,0.79 1.00).The diagnostic efficacy of this model in the diagnosis of lymphatic metastasis was higher than the RAD score and clinical parameters,with AUC(0.90 vs 0.86 vs 0.68)in the training cohort and(0.91 vs 0.90 vs 0.54)in the validation cohort.Boot was used to verify this Nomogram in patients of the training cohort and the validation cohort.2000 repeated tests in the training cohort and the validation cohort showed that the diagnostic efficacy of the Nomogram was within the 95%CI interval,and the difference was statistically significant(P=0.006).2.The Rad score was constructedbased on tumor image segmentation of mpMRI and radiomics parameters of patients with cervical cancer to evaluate whether lymphatic cancer metastasis exits in patients recived surgery only.1).The Rad score was constructed based on T2WI segmentation of cervical tumors to evaluate lymph node metastasis The AUCwas 0.74(95%CI,0.63-0.86)in the training cohort and 0.69(95%CI,0.52-0.85)in the validation cohort.The accuracy,sensitivity and specificity were 0.727,0.469 and 0.833 in the training cohort,and 0.577,0.277 and 0.778 in the validation cohort,respectively.2).The Rad score was constructed based on T1WI enhanced arterial phase segmentation of cervical tumors to evaluate lymph node metastasis,the AUC was 0.79(95%CI,0.70-0.88)in the training cohort and 0.70(95%CI,0.52-0.89)in the validation cohort.The accuracy,sensitivity and specificity were 0.718,0.465 and 0.880 in the training cohort,and 0.533,0.307 and 0.842 in the validation cohort,respectively3).The Rad score was constructed based on T1WI enhanced venous segment of cervical tumors to evaluate lymph node metastasis the AUCwas 0.77(95%CI,0.66-0.88)in the training cohort and 0.63(95%CI,0.41-0.84)in the validation cohort.The accuracy,sensitivity and specificity were 0.818,0.900 and 0.810 in the training cohort,and 0.755,0.500 and 0.767 in the validation cohort,respectively.4).The multivariate logistic regression model Nomogram visualization was developed by using the image segmentation radiomics tag Rad score based on the optimal diagnostic performance T1WI enhanced arterial cervical tumor in combination with clinical parameters(preoperative MRI measuring tumor diameter line and largest number of pregnancy),wich was used for the assessment of lymph node metastasis,The AUC in training cohort was 0.83(95%CI,0.74 0.92),and in the validation cohort was 0.70(95%CI,0.51 0.88).The diagnostic efficacy of this model for lymphatic metastasis was higher than that of the Rad score and clinical parameters,with AUC(0.83 vs 0.79 vs 0.70)in the training group and(0.70 vs 0.70 vs 0.56)in the validation group,respectively.Conclusion:Nomogram visualization of multiple logistic regression model based on T1WI enhanced arterial-segment cervical tumors with optimal pre-treatment diagnostic efficacy and clinical parameters can be used to predict lymph node metastasis of patients with cervical cancer treated with neoadjuvant chemotherapy combined with surgery or surgery only in an early and non-invasive way.Part ⅢPredicting postoperative progression-free survival of patients with cervical cancer based on mpMRI raiomics combined with clinical parametersObjective:By removing redundancy and impregnation of Radiomics features extracted from cervical tumors by mpMRI before treatment,Rad score combined with clinical parameter modeling were constructed to predict progression-free survival of cervical cancer patients without invasion.Methods:Collection in the second part of this research into the group of neoadjuvant chemotherapy in patients with clinical and MRI imaging parameters,including age,body mass index,FIGO staging before treatment,neutrophils,absolute value of absolute value of lymphocyte,monocyte absolute value,platelet,hemoglobin,serum albumin,SCC-Ag,CEA,CA125,HPV,MRI measured before treatment tumor maximum diameter line,the minimum ADC values,etc.The end point of this study was postoperative recurrence or metastasis.Patients were randomly divided into training group and validation group according to 7:3.With the second part studies from the cervical axial T2WI,transverse T1WI enhancement cervical tumor arterial and venous phase sequence images to extract the Radiomics characteristics,first based on a single MRI sequence image respectively Rad score build,predicting progression-free surial postoperatively in patients with cervical cancer,with clinical parameters build multivariate COX model survival risk model,and use the Nomogram visualization predicting progression-free surial postoperatively in patients with cervical cancer,3,4,and 5 years survival rate of patient evaluation.Log rank analysis was performed based on clinical univariate analysis of patients in the training group,and parameters of HR>1 were included in Cox model.C index and IAUC were used to evaluate the prediction performance of the model in the training set and the verification set respectively.Results:A total of 82 patients with neoadjuvant chemotherapy were enrolled,of which 18 patients had recurrence and 64 patients had no recurrence.1.Rad score was constructed based on T2WI segmentation of cervical tumors with recurrence or metastasis as the classification label.The minimum penalty coefficient LoG λ=0.049 was used to construct the RAD score of the Rad score corresponding to seven final imaging features.The Rad score in the training group and the test group was analyzed by univariate KM analysis according to the patients’ PFS.The log rank test(P<0.0001)of the RAD score in the training group and the log rank test(P=0.96)of the validation group was performed.The diagnostic efficacy of the RAD score in the training group and the validation group was evaluated by C-index,which was 0.849 in the training group and 0.711 in the validation group.2.Rad score was constructed based on the classification label of recurrence or metastasis of cervical tumors in T1WI enhanced venous phase segmentation.The minimum penalty coefficient Log λ=0.028 was used to construct the Rad score for 9 corresponding Radiomics features.The Rad score in the training group and the test group was analyzed by univariate KM analysis according to the patients’ PFS.The Rad score was P<0.0001 in the training group and P=0.48 in the validation group.The diagnostic efficacy of the Rad score in the training group and the validation group was evaluated by C-index,which was 0.828 in the training group and 0.697 in the validation group.3.Rad score was constructed based on the classification label of recurrence or metastasis of cervical tumors in T1WI enhanced arterial phase segmentation.The Rad score was constructed by taking the minimum penalty coefficient LoGλ=0.132 and two final corresponding imaging omics features.The Rad score in the training group and the test group was analyzed by unifactorial KM according to the patients’ PFS.The Rad score in the training group was tested by Log Rank test(P=0.00081),and in the validation group by Log Rank test(P=0.0026).The diagnostic efficacy of Rad score in the training group and the validation group was evaluated by C-index,which was 0.896 in the training group and 0.802 in the validation group.4.Radiomics signature were constructed based on T1WI enhanced arterial magnetic resonance imaging combined with clinical parameters(preoperative platelet,neutrophils)to construct a Cox model to evaluate the progression-free survival of cervical cancer patients.The C-index of Nomogram model for diagnostic efficacy training group was 0.948,and the C-index of validation group was 0.849,indicating a good fit with the ideal model.The efficacy of progression-free survival was predicted by the IAUC evaluation model,with the IAUC of 0.943 in the training group and 0.865 in the test group.Conclusion:Based on T1WI enhancement sequence arterial cervical cancer before treatment to extract the Rad score joint clinical parameters(preoperative platelet,neutrophils)build multivariate logistic regression model Nomogram visualization,no invasive early predicting progression-free surial neoadjuvant chemotherapy of cervical cancer patients,help clinical setting individualized treatment plan. |
PDF Full Text Request