| Purpose:Soft Tissue Tumors（STTs）are highly heterogeneous,and malignant STTs often have the risk of invasion,recurrence and metastasis.Accurate preoperative diagnosis can avoid unplanned resection of malignant STTs and excessive treatment of nonmalignant STTs.However,for soft tissue sarcomas（STSs）,it is difficult to predict the clinical course only by providing the diagnosis of malignant STTs,and grade evaluation is needed.The long-term prognosis and clinical treatment strategy of STSs in different grades vary greatly.US and MRI are two high-resolution methods,were widely used in malignant STTs diagnosis.Previous studies focused on the independent analysis of the two,but there were few reports on their combined application.The purpose of this study was trying to build US and MRI joint nomogram for diagnosis of malignant STTs and STSs grading,through analyzing the risk factors of the two modes.If the construction of the combined diagnostic models fails,only the single dominant mode will be used to construct the nomogram,and the radiomics will be combined to construct the radiomics nomogram,and the model with the highest diagnostic performance will be selected.Materials and methods:1.From September 2016 to December 2020,134 STTs patients of head and neck,trunk or limbs were first diagnosed in our hospital,including 71 cases of malignant and 63 cases of non-malignant,including 103 cases of MRI,59 cases of malignant and 44 cases of non-malignant,91 cases of US image,42 cases of malignant and 49 cases of non-malignant.A total of 60 cases underwent US and MRI examination at the same time.31 cases were malignant and 29 cases were non-malignant.The routine US signs,elastic images,routine MRI signs,DWI signs and clinical factors were analyzed.The clinical factors,US and MRI signs parameters of 134 patients were analyzed by univariate analysis and multi-factor logistic regression analysis.The parameters with independent predictive value were screened out respectively.The above multivariate parameters were analyzed again in the cohort of 60 patients who underwent both US and MRI examinations,and the parameters with independent predictive value for malignant STTs were screened to construct the nomogram and the diagnostic efficacy of the nomogram models based on US,MRI and US+MRI factors for predicting malignant STTs were compared.2.Patients pathologically diagnosed with STSs in our hospital from January 2015 to December 2020,using the STSs grading criteria of the French Federation of Cancer Centers（FNCLCC）,patients in each group were divided into two groups: high grade（Ⅲ）and low grade（Ⅰ,Ⅱ）.There were a total of 79 cases,40 cases of high grade and 39 cases of low grade,including 54 cases of US image,24 cases of high grade,30 cases of low grade,including 19 cases of SE image,6 cases of high grade and 13 cases of low grade,57 cases of conventional MRI and DWI image,29 cases of high grade and 28 cases of low grade.The routine US signs,elastic images,routine MRI signs,DWI signs and clinical factors were analyzed.Univariate analysis and multifactor logistic regression analysis were conducted for clinical factors,US,SE and MRI signs parameters.The nomogram was based on the queue which could screen out the independent predictive value parameters.The ROC curve and calibration curve of the nomogram were used to evaluate the diagnostic performance and calibration degree.3.The T2WI-FS sequence images of 57 MRI images were imported into the radiomics cloud platform in the DICOM format,and the ROI of tumor was manually delineated layer by layer,and the focus VOI was automatically generated by computer.Set the tumor body outward expansion width of 5mm on the platform to obtain the original VOI of peritumoral area.The actual VOI of peritumoral area was obtained by subtracting the air part around the tumor.The feature was screened by dimensionality reduction in four steps: minimum-maximum normalization,optimal feature percentage,selection according to SVM-L1 model,and iterative feature screening.Radiomics characteristics of 57 patients after screening were divided into training and validation group（8:2）randomly.The AUC,95% CI,accuracy,sensitivity specificity and AUC after 5 fold cross-validation in support vector machine（SVM）were calculated in the T2WI-FS intratumoral,peritumoral and combination group.Radiomics nomogram was constructed by Rad-score based on the best radiomics model with combination of MRI signs T2 WI signal homogeneity,and tumor peritumoral enhancement.Radiomics nomogram AUC value,95%CI,accuracy and sensitivity,and specificity were calculated.The diagnostic efficiency of the radiomics nomogram,radiomics model and the MRI sign nomogram were compared.Result:1.US signs: the echo homogeneity,edge,blood flow type,ES and EI/B were statistically different between the two groups.The results of multivariate analysis showed that irregular edge morphology and EI/B value >1 were independent predictors of malignant STTs.There were statistical differences between the two groups in T1 WI margin,T2 WI margin,T2 WI signal homogeneity,T2 WI peritumoral edema,T2 WI tail sign,neurovascular bundle involvement,and limited DWI visual dispersion.In multivariate analysis,heterogeneous signal intensities greater than 50% at T2 WI and DWI diffusion greater than 50% were independent predictors of malignant STTs.There were statistical differences in gender,age,RALD and depth between the two groups.In the results of multivariate analysis,age,RALD and across fascia were independent predictors of malignant STTs.Age,RALD>0.5,heterogeneous signal intensities greater than 50% at T2 WI,and irregular edge of US were independent risk factors for predicting malignant STTs.The diagnostic efficiency of age +US+MRI nomogram was better than that of age +MRI or age +US.The AUC value（95%CI）of the nomogram based on age,RALD,T2 WI signal homogeneity,and US edge in the training set and validation set（7:3）were 0.97（0.93,1）and 0.96（0.84,1）,respectively,and the accuracy rates were 88.81% and 88.89%,respectively.2.There were statistically significant differences in echo heterogeneity,blood type,ES and EI/B between the high and low grade groups,while multivariate analysis did not screen out any factors with independent predictive value for high grade STSs.There were statistically significant differences in RALD,T2 WI signal homogeneity,T2 WI peritumoral edema,the presence of tumor peritumoral enhancement between the high and low grade groups.In the multi-factor analysis results,heterogeneous signal at T2-weighted imaging and peritumoral enhancement were independent risk factors for prediction of high grade STSs.The nomogram was constructed based on T2 WI signal homogeneity,and tumor peritumoral enhancement.The AUC（95%CI）of diagnosis of high grade STSs in the training set and the validation set（7:3）were 0.86（0.7,1）and 0.83（0.51,1）,respectively,and the accuracy rates were 82.05% and 72.22%,respectively.3.7 features were selected in intratumoral group,the verification AUC=0.72（0.33,1）and the five-fold cross-validation AUC=0.78.The accuracy,sensitivity and specificity of validation group were 58.33%,83.33% and 66.67%,respectively.6 features were selected in peritumoral group,the verification AUC=0.92（0.72,1）and the five-fold cross-validation AUC=0.93.The accuracy,sensitivity and specificity of validation group were 75%,83.33% and 83.33%,respectively.5 features were selected in intratumoral + peritumoral combination group,the verification AUC=0.86（0.45,1）and the five-fold cross-validation AUC=0.93.The accuracy,sensitivity and specificity of validation group were 91.1%,100% and 83.33%,respectively.The radiomics nomogram was constructed by combining T2 WI signal homogeneity,peritumor enhancement and radiomics Rad-score.The AUC（95%CI）of the training and validation group for predicting high grade STS were 0.99（0.97,1）and 0.96（0.75,1）,respectively.The accuracy,sensitivity and specificity were 94.87%,100%,94.7% and 94.4%,100% and 88.9%,respectively.The calibration curve is attached to the central 45-degree diagonal,indicating that the diagnostic efficiency of the nomogram model is stable.Conclusion:1.The irregular edge of US,EI/B>1,heterogeneous signal intensities greater than 50% at T2 WI and DWI diffusion greater than 50% are important indicators for the diagnosis of malignant STTs.The combined diagnostic performance of US and MRI is better than that of the single model constructed by them.The model based on US edge,T2 WI signal heterogeneity,RALD and age has superior diagnostic performance for the prediction of malignant STTs.2.In the diagnosis of STSs grading,T2 WI signal homogeneity,and tumor peritumoral enhancement in MRI signs are important indicators for the diagnosis of high-grade STSs.The MRI sign nomogram based on these two signs and the intratumor + peritumor imaging model based on T2WI-FS both have good diagnostic performance.The radiomics nomogram constructed from the Rad-score and MRI signs performed best in the diagnosis of high-grade STSs. |
