[Objectives]To construct and validate a nomogram model based on axial T2-weighted imaging(T2WI)and diffusion-weighted imaging(DWI)radiomics features and related clinical factors,and to explore the clinical application value of this model in differentiating uterine sarcoma(US)from atypical leiomyoma(ALM)preoperatively.[Materials and Methods]From July 1,2015 to June 1,2022,in Qilu Hospital of Shandong University,62 patients with US and 77 patients with ALM who were confirmed by postoperative pathology and underwent pelvic magnetic resonance imaging(MRI)examination prior to surgery were retrospectively collected,including clinical,pathological and imaging data,among whom 43 patients with US and 53 patients with ALM were enrolled in the training cohort,and the other 19 patients with US and 24 patients with ALM were enrolled in the test cohort according to the ratio of 7:3.Pathological examination was considered as the gold standard of differentiating US from ALM.Univariate and multivariate logistic regression analyses were used to analyze and select the independent risk factors of clinical characteristics(age,menopausal status,abdominal pain,abdominal distension,abnormal vaginal bleeding,menstrual change and palpable mass)and conventional MRI characteristics(tumor maximum diameter,margin,shape,hemorrhage,necrosis or cystic areas,flow voids,pelvic effusion and pelvic lymph node status)to distinguish US from ALM,and the clinical model was established.The region of interest(ROI)on each slice of the entire tumor was manually delineated along the tumor outline on axial T2WI and DWI sequences using ITK-SNAP software,and the radiomics features of ROIs were extracted with uniform imaging following tumor segmentation.Then,variance threshold,Select K Best and least absolute shrinkage and selection operator(LASSO)regression were used to screen radiomics features,construct the T2WI+DWI radiomics signature and calculate the radiomics score(Rad-score).The combined model was constructed by combining independent risk clinical factors(clinical and conventional MRI characteristics)and Rad-score,and the radiomics nomogram was drawn.The receiver operating characteristic curve(ROC)was used to analyze the area under the curve(AUC)to evaluate the diagnostic performance of three models(clinical model,T2WI+DWI radiomics signature and nomogram model)in the training and test cohorts.The specificity,sensitivity and accuracy of the three models were calculated.Delong test was used to analyze the differences of AUCs between any two models of three models.Decision curve analysis(DCA)was used to evaluate the clinical application value of the nomogram model.[Results](1)The results of univariate and multivariate logistic regression analyses showed that abnormal vaginal bleeding(OR=9.831,95%confidence interval 2.299 to 42.038,P=0.002)and tumor shape(OR=13.258,95%confidence interval 3.374 to 52.088,P<0.001)were two independent risk clinical factors in the differential diagnosis of US and ALM.The AUCs of the clinical model in the training and test cohorts were 0.879 and 0.822,respectively.(2)2818 radiomics features were extracted from the axial T2WI and DWI sequences.Finally,13 important and robust radiomics features were selected to build the radiomics signature and calculate the Rad-score,among which 8 radiomics features were from axial T2WI sequences and 5 radiomics features were from axial DWI sequences.The AUCs of the radiomics signature based on T2WI+DWI sequences in the training and test cohorts were 0.912 and 0.842,respectively.(3)The nomogram model was constructed by combining independent risk clinical factors and Rad-score.The AUCs of the nomogram model in the training and test cohorts were 0.965 and 0.958,respectively.The calibration curve showed that the nomogram model had high reliability.(4)The AUCs of the clinical model and radiomics signature were lower than that of the nomogram model(training cohort:Z=-3.394,P<0.001 and Z=-2.121,P=0.034;test cohort:Z=-2.874,P=0.004 and Z=-2.4089,P=0.016).However,there was no significant difference in AUCs between the clinical model and radiomics signature based on T2WI+DWI sequences(training cohort:Z=-0.760,P=0.448;test cohort:Z=-0.237,P=0.813).The results of DCA showed that the overall benefit of the nomogram model was higher than that of the clinical model and radiomics signature based on T2WI+DWI sequences.[Conclusions]Abnormal vaginal bleeding and tumor shape are meaningful in distinguishing US from ALM.Radiomics also has potential value in the differential diagnosis of these two tumors.The nomogram model combined with independent risk factors of clinical and conventional MRI characteristics and Rad-score can improve the differential diagnostic performance effectively,and can help to make precise diagnosis and clinical decision. |