| Objective:1 Endometrial carcinoma is a common malignant tumor in women.Although the prognosis is good,as many as 15-20%of the patients have recurrence,and the survival rate after recurrence is low.Lymph node metastasis is a poor prognostic indicator of endometrial cancer,but lymph node dissection can not reduce the risk of recurrence and death in patients with early endometrial cancer.Therefore,the preoperative non-invasive and convenient evaluation of lymph node metastasis is very important.The purpose of this study is to combine clinical and preoperative PET/CT imaging information to construct a model for predicting lymph node metastasis of endometrial carcinoma.2 The purpose of this study is to combine clinical physicochemical information and PET/CT radiomics features to construct a model for predicting 3-year and 5-year survival of endometrial cancer.Methods:Part I1.1 Patients with endometrial cancer who completed PET/CT in Shengjing Hospital of China Medical University and underwent double adnexal hysterectomy and lymph node dissection（pelvic±aortic side）were retrospectively collected from July 1,2012 to December 31,2018.1.2 Clinical information(age,menopausal status,CA125 level,maximum tumor diameter,18F-FDGPET/CT suggested lymph node status,pathological diagnosis and histological differentiation,postoperative lymph node metastasis status)were collected.Logistic regression model was used to screen the clinical information related to postoperative lymph node metastasis.1.3 Extract PET/CT images,draw regions of interest,extract tumor metabolic parameters（SUVmax,SUVmean,MTV,TLG）and histological features.Logistic regression model was used to evaluate the relationship between the metabolic information and postoperative lymph node metastasis.Logistic regression model was used to explore the associations between PET and CT radiomics features and postoperative lymph node metastasis.LASSO regression and cross-validation were used to select the best PET and CT radiomics features and to construct PET and CT radiomics features labels.1.4 Four models（clinical model,clinical+metabolic parameter model,clinical+PET radiomics features model,clinical+CT radiomics features model）were conducted.The area under the working curve of the different models was calculated,the calibration curve was drawn,and the optimal model was selected.1.5 Nomogram was constructed to predict postoperative lymph node metastasis of endometrial carcinoma by using the optimal model.Part II2.1 Information of patients with endometrial cancer who were treated in Shengjing Hospital of China Medical University from July 1,2012 to December 31,2018 and completed PET/CT before treatments were retrospectively collected and selected by inclusion and exclusion criteria.2.2 Clinical information was collected,including age,menopausal status,serum CA125level,maximum tumor diameter,pathological diagnosis and histological differentiation,lymph node metastasis,depth of myometrial invasion,lymphatic vascular space infiltration and cervical interstitial involvement.COX regression model was used to evaluate the relationship between clinical information and survival.2.3 The PET/CT images were extracted,and the regions of interest were delineated.The metabolic parameters（SUVmax,SUVmean,MTV,TLG）were extracted from the regions of interest.COX regression model was used to evaluate the associations between metabolic information and survival time.PET and CT radiomics characteristics were firstly selected by COX regression model,and then LASSO regression and cross validation were used to select the best PET and CT radiomics features and construct PET and CT radiomics labels.The survival curves were conducted to determine the correlation between the established label and prognosis.2.4 Construction model:The information related to prognosis were selected to construct the model respectively.The consistency indexes of different models were calculated,the calibration curves were drawn,and the optimal model was selected.2.5 Nomogram was constructed to predict the 3-and 5-year survival time of endometrial cancer by using the optimal model.Results:Part I1.1 Following inclusion and exclusion criteria,311 patients with endometrial carcinoma were randomly divided into training set（n=221）and test set（n=90）at 7:3.There was no difference in general characteristic between two groups.1.2 Univariate Logistic regression model showed that CA125,maximum tumor diameter and high metabolism of PET were correlated with lymph node metastasis.Univariate Logistic regression model showed that Volume and TLG were associated with lymph node metastasis.Whenλ=0.0190,24 features were selected from PET radiomics features to construct PET radiomics labels.Whenλ=0.0036,7 features were selected from CT radiomics features to construct CT radiomics labels.The AUC of PET+clinical model in the training set was 0.916（95%confidence interval 0.867-0.963,sensitivity:0.762,specificity:0.939）,which was higher than those of other models.The AUC of PET+clinical model in the test set was 0.796（95%CI:0.629-0.904,sensitivity:0.722,specificity:0.764）,which was higher than those of other models.1.5 Nomograph was constructed based on clinical information and PET radiomics labels to predict postoperative lymph node metastasis of endometrial carcinoma.Part II2.1 A total of 311 patients with endometrial cancer were randomly assigned to the training set（n=224）and the test set（n=87）at 7:3.There was no difference in general characteristic between two groups.2.2 Univariate COX regression model showed that serum CA125 level,lymph node metastasis,myometrial invasion and lymphatic vascular space infiltration were correlated with overall survival（P<0.05）.2.3 The metabolic parameters were not related to the overall survival time in univariate COX regression model（P>0.05）.Whenλ=0.0140,8 features were selected from PET radiomics features to construct PET radiomics labels.Whenλ=0.0021,7 features were selected from CT radiomics features to construct CT radiomics labels.The low classification of PET labels and the low classification of CT labels suggest a good prognosis.2.4 Three models（clinical model,clinical+PET labels model and clinical+CT labels model）were conducted.The consistency index of clinical+PET label model in training set was 0.88（95%CI:0.85-0.91）,and that in test set was 0.84（95%CI:0.74-0.92）,both of which were better than other models.2.5 Nomogram was constructed to predict the 3-year and 5-year survival of endometrial cancer based on clinical information and PET radiomics labels.Conclusion:1 By reviewing the preoperative PET/CT images of 311 patients with endometrial carcinoma diagnosed by Shengjing Hospital of China Medical University,we selected the most effective PET features for predicting lymph node metastasis by extracting,screening and comparing them,and constructed a model for predicting lymph node metastasis by integrating PET radiomics labels with clinical information.A noninvasive PET based nomogram was constructed for the first time.It is effective in predicting lymph node metastasis,and provide individualized treatment guidance.Its clinical application value needs to be further verified by more prospective multicenter studies.2 The imaging feature labels extracted from PET and CT images of PET/CT images can quantitatively characterize intra-tumor heterogeneity and were related to the overall survival of patients with endometrial carcinoma.Compared with the model using clinical features alone or in combination with CT imaging feature labels,clinical+PET radiomic labels models were more accurate and effective in predicting the overall survival of patients with endometrial cancer.The Nomogram based on this model is an effective tool for predicting the 3-year and 5-year survival of patients with endometrial cancer. |
PDF Full Text Request