Radiomics Analysis Of ¹⁸F-FDG PET/CT Imaging For Prediction Of Lymph Node Metastasis In Epithelial Ovarian Cancer And Progression-free Survival In High Grade Serous Carcinoma

Background: Ovarian carcinoma is the leading cause of gynecologic cancer deaths,because majority of patients were diagnosed with advanced-stage disease(stages Ⅲ and Ⅳ)according to International Federation of Gynecology and Obstetrics(FIGO)staging classification.In patients with stage Ⅲ-Ⅳ ovarian cancer,lymph node metastasis was more than 50%.There was no significant correlation between lymph node metastasis and the size of lymph node,which reduced the sensitivity of PET / CT in detecting lymph node metastasis.LNM is an important prognostic factor in patients with epithelial ovarian cancer.High-grade serous ovarian cancer(HGSOC)accounts for up to 70% of women with epithelial ovarian carcinoma.Although most of patients with HGSOC achieve complete remission with the cytoreductive surgery and cisplatin-based chemotherapy,the median progress-free survival(PFS)is only 18 months.A significant proportion of advanced HGSOC patients experience tumor recurrence and progress within 3 years.Identification of recurrence in patients with advanced HGSOC is important since it guides the decisions regarding personalized treatment and surveillance planning.Radiomics based on high-dimensional quantitative features extracted from different medical imaging modality can noninvasively quantify tumor heterogeneity and show underlying malignant features.On the basis of prediction model based on those radiomics features,clinicians can deliver more personalized medical care with regard to tumor diagnosis,histopathology classification,assessment of therapeutic responses and prognosis.Therefore,we established radiomics of PET and CT,and hybrid radiomics nomograms integrating RS and clinical factors for prediction of lymph node metastasis and progress-free survival.In addition,the performances of these hybrid nomograms were compared.Objective: This study aims to investigate radiomics features extracted from PET and CT components of 18 F-FDG PET/CT images integrating clinical factors,to predict LNM in epithelial ovarian cancer and progression-free survival in HGSOC.Methods: In part 1,a total of 275 patients were finally enrolled in the study and randomly divided into training(n=189)and validation cohorts(n=86).The ROIs were drawn by two experienced radiologists.The radiomics features of PET and CT images were extracted by using a module called pyradiomics.Lasso regression analysis were used to construct the radiomics signatures of PET(PET＿RS)and CT(CT＿RS)respectively with the features with P value <0.1 in univariate analysis.Three model were constructed by multivariate logistic regression analysis(clinical,clinical+PET＿RS,and clinical+CT＿RS).Models were evaluated by the receiver operating curves,calibration curves,and net reclassification index.The best model was constructed into a nomogram.In part 2,A total of 265 patients were finally enrolled in the study and randomly divided into training(n=180)and validation cohorts(n=85).All VOIs of PET/CT images were semiautomatically segmented with threshold segmentation of 42% of maximal SUV in PET images.A total of 850 features were separately extracted from PET and CT components of 18 F-FDG PET/CT images,and two radiomics signatures(RSs)were then constructed by least absolute shrinkage and selection operator(LASSO)method.Clinical features and metabolic parameters of PET(PET＿MP)with P value <0.05 in univariate analysis was combined with RS of PET(PET＿RS)and CT components(CT＿RS)to develop six prediction nomograms(Clinical,Clinical+ PET＿MP,Clinical+ PET＿RS,Clinical+ CT＿RS,Clinical+ PET＿MP + PET＿RS,Clinical+ PET＿MP + CT＿RS)using multivariate COX regression.The concordance index(C-index),NRI,and calibration curves were applied to evaluate the performance of nomogram in training and validation cohorts.Results: In part 1,the clinical+CT＿RS model had best predictive performance in train cohort.The AUC,sensitivity,and specificity of this model was 0.710(95% CI0.597-0.817),0.886,and 0.452 in the training cohorts.But in validation cohorts,the clinical+PET＿RS model had best predictive performance.The AUC,sensitivity,and specificity was 0.713(95% CI 0.602-0.825),0.750,and 0.652 in the validation cohorts.The clinical+PET＿RS model show good calibration in both training and validation cohorts in calibration plots.we constructed a nomogram for clinical+PET＿RS model.In part 2,two PET radiomics features was selected by LASSO to construct PET＿RS,and 13 CT radiomics features to construct CT＿RS.There was significate difference between low and high RS groups for PET＿RS in training(P<0.001)and validation cohorts(P=0.003),and for CT＿RS in train(P<0.001)and validation cohorts(P=0.004).Clinical+ PET＿MS + CT＿RS model showed higher prognostic performance than other models(C-index=0.72)in training cohorts,while the Clinical+ PET＿MS +PET＿RS model had highest prognostic performance in validation cohort(C-index=0.681).Calibration curves of each model for prediction of 1-,3-year PFS showed Clinical +PET＿MP + PET＿RS model show excellent agreements between estimated and the observed 1-,3-outcomes.Compared to basic clinical model,Clinical+ PET＿MS +PET＿R model resulted in greater improvement in prediction performance.Conclusion: PET＿RS can improve diagnostic accuracy and provide complementary prognostic information compared with use of clinical parameters alone or combined with CT＿RS.The newly developed radiomics nomogram is an effective tool to predict lymph node metastasis in epithelial ovarian cancer and PFS in HGSOC.