Research On Preoperative Prediction Of Uterine Sarcoma By A Combined Model Based On MRI Radiomics And Clinical Variables

Objective:The purpose of this study was to develop and validate a diagnostic model based on T2+DWI radiomics and clinical variables for preoperative differentiation of uterine sarcoma and uterine fibroids.Materials and methods:This retrospective study included 99 patients(28 patients with uterine sarcomas and 71 patients with uterine fibroids)who underwent pelvic MRI imaging preoperatively from July 2013 to December 2021 in the Affiliated Hospital of Qingdao University.The examinations of all patients were performed on our institution’s GE 3.0T MRI.The imaging sequences included axial fast-spin-echo(FSE)T2WI fat-suppressed(FS)sequence and diffusion weighted imaging(DWI)sequence.Patients were randomly assigned to training set(69 cases)and validation set(30cases).IBM SPSS software was used to statistically analyze the clinical characteristics(age,menopausal status,hemoglobin count,neutrophil count and LDH count etc.)of patients with uterine sarcomas and uterine fibroids.ROC analysis of clinical variables is performed using the p ROC package in R and taking the corresponding cut-off value as the boundary value of each clinical variables.Incorporating age,BMI,NLR,PLT,LDH and other clinical variables into univariate and multivariate logistic regression,screening out independent preoperative risk factors,and then building a clinical model.Using ITK-SNAP software,all levels of the largest lesions(when there are multiple lesions)in each patient’s T2 WI and DWI sequences were manually outlined to serve as regions of interest(ROI).This study used the opensource Python Py Radiomics package for radiomics to extract the features of ROI.Using the least absolute shrinkage and selection operator(LASSO)and one-way analysis of variance(ANOVA)to screen the extracted radiomics features,we constructed image labels and subsequently established a radiomics model based on T2+DWI sequences through logistic regression algorithm.The clinical-radiomics model——nomogram was constructed by combining the radiomics model with meaningful clinical variables and the calibration curves were used to evaluate the calibration of the nomogram.The goodness of fit of the nomogram was evaluated by the Hosmer-Lemeshow test.The diagnostic effects of the radiomics model,clinical model,and combined clinical-radiomics model were compared by the area under the receiver operating characteristic curve(AUC),accuracy,sensitivity,and specificity.Delong test was used to evaluate the difference of AUC between any two combinations of the three models,and the net income of the three models was drawn through decision curve analysis(DCA)to evaluate the clinical usefulness.Finally,the diagnostic performance was compared with that of clinical physicians in the whole patient cohort.Results:Univariate and multivariate logistic regression analysis confirmed that LDH and NLR were independent risk factors for predicting uterine sarcoma.By outlining and feature extraction and screening of ROI in both T2 WI and DWI sequences,16 radiomics features were finally obtained for the construction of radiomics model,including 7 features from T2 WI and 9 features from DWI.In the training set,the combined model(nomogram)had an improved AUC value(0.982±0.024)compared to the T2WI+DWI radiomics(AUC: 0.977±0.031)and the clinical model(AUC:0.756±0.122);in the validation set,the nomogram had an improved AUC value(0.903±0.124)compared to the radiomics(AUC:0.835±0.171)and the clinical model(AUC:0.869±0.178).The nomogram had good calibration and clinical net benefit.For the identification of uterine sarcoma,the nomogram had higher sensitivity than two clinical physicians(0.9 vs 0.714).Conclusion:The radiomics model based on T2WI+DWI had good performance in distinguishing uterine fibroids from uterine sarcomas,and the combined model(nomogram)had the best diagnostic performance among the three models,which was better than the diagnostic performance of clinical physicians and could provide effective reference for clinical decision-making.