| Objectives: To investigate the predictive value of the radiomics model in pelvic lymph node metastasis(PLNM)of patients with early-stage cervical squamous cell carcinoma(CSCC)based on the apparent diffusion coefficient(ADC)maps.Methods: A total of 153 patients with early-stage CSCC were examined with routine MRI including DWI preoperatively at the Liaoning Cancer Hospital&Institute(Liaoning Cancer Center,Dalian Medical University Clinical Oncology College)from January 2015 to October 2017.All the patients were divided into two groups with 2:1ratio by time sequence: training group(n=102)and validation group(n=51).The data of patients’ age and clinical FIGO stage were obtained by reviewing patient medical record.Tumour diameter was defined as the maximal tumour diameter measured on T2-weighted images in the sagittal,coronal,and transverse axial planes;and presence of MR-reported LN-positive metastases was defined as larger than 10 mm in the maximal short-axis diameter that was measured on T2-weighted in transverse axial planes.The ADC maps of all patients were segmented layer by layer to form VOI,and the entire lesion was included as far as possible along the edge of the lesion.Radiomic feature extraction was performed using the texture extraction software,and a total of66 radiomics parameter features were extracted for each patient.Among the 66 texture features,the one-way analysis of variance(ANOVA)method and Mann-Whitney U test were used to select the PLNM-related predictive features in the training data set,and the Pearson’s correlation analysis was used for redundantprocessing.Independent sample t-tests was used to compare the age difference between LN metastasis group and non-metastasis group;Chi-square test was used to compare the differences in clinical stage,tumour diameter and MR-reported LN status between the two groups.A multivariate logistic regression analysis was carried out on the PLNM-related predictors to establish a pre-predictive model of PLNM in patients with early-stage CSCC,and using Receiver operating characteristic curve(ROC)to evaluate the performance of the radiomics model as well as predictors in the training group.In the validation group,the difference in patient age between LN metastasis group and non-metastasis group was compared using an independent sample t-test;the differences in clinical stage,tumour diameter and MR-reported LN status between the two groups were compared using a Chi-square test.The relevant clinical features,radiological features and texture features of the validation group were substituted into the prediction model derived from the training group,and using ROC curve to evaluate the performance of the radiomics model as well as predictors.Result:1.Establishment of the Radiomics model(Training group):1)Among the 102 patients,the patient age in LN metastasis group was slightly lower than that of non-metastasis group,which was no significant difference(P>0.05).There were statistically significant differences in the clinical stage,tumour diameter and MR-reported LN status(P<0.05)between LN metastasis group and non-metastasis group.2)Of the 66 radiomic features,max intensity,cluster prominence,inverse difference moment,grey-level non-uniformity were the most significant predictive characteristics for early-stage CSCC patients.The max intensity,inverse difference moment and grey-level non-uniformity in LN metastasis group was slightly higher than those of non-metastasis group,and cluster prominence was slightly lower than that of non-metastasis group,which were statistically significant(P<0.05).3)Based on the clinical-radiological features and radiomics features of whole-lesion ADC maps,a preoperative PLNM prediction model for early-stage CSCC patients was established,which including the clinical stage,MR-reported LN status and grey-level non-uniformity.The critical value of the model was 0.113,which meant a patient with an estimated probability greater than 11.3% was considered to have PLNM.4)The area under the curve(AUC)of the radiomics model was 0.864,which was higher than those of the clinical stage(AUC,0.607),tumour diameter(AUC,0.668)and MR-reported LN status(AUC,0.664).The diagnostic accuracy,sensitivity,specificity,positive predictive value and negative predictive value of the radiomics model were 0.804,0.846,0.778,0.702 and 0.891 respectively.2.Validation of the Radiomics model(validation group):1)There were no significant differences in the clinical stage,tumour diameter and MR-reported LN status between the training and validation cohorts(P>0.05),indicating that these two groups of patients could be used to develop and validate the model.2)Among the 51 patients,there was no significant difference in patient age between LN metastasis group and non-metastasis group,while there were statistically significant differences in the clinical stage,tumour diameter and MR-reported LN status(P<0.05).3)Compared with clinical stage(AUC,0.744),tumour diameter(AUC,0.744)and MR-reported LN status(AUC,0.772),the AUC of the radiomics model was the highest(AUC,0.870).The diagnostic accuracy,sensitivity,specificity,positive predictive value and negative predictive value of the radiomics model were 0.784,0.867,0.750,0.591 and 0.931 respectively.Conclusion:1)Clinical stage,tumour diameter and MR-reported LN status are the predictors of PLNM in early-stage CSCC patients.2)The radiomics model based on the whole-lesion ADC maps included the clinical stage,MR-reported LN status and grey-level non-uniformity.The critical value of the model was 0.113,which meant a patient with an estimated probability greater than 11.3% was considered to have PLNM.3)Compared with a separate study of clinical or radiological features,the radiomics model showed good predictive efficacy in both the training group and the validation group.It can be used as a non-invasive predictive tool to preoperatively evaluate the pelvic lymph node status of early-stage CSCC patients. |
PDF Full Text Request