The Value Of MRI Radiomic In Differentiating Benign And Malignant Soft Tissue Masses With Cystic-appearing

Objective:To investigate the feasibility and application value of imaging radiomic models based on T₂WI-FS、T₁WI-FS and T₁WI-FS+C sequences in differentiating benign and malignant cystic-appearing soft tissue masses.Materials and Methods:1.Clinical data Retrospective analysis was performed on patients who underwent MRI examination the Second Hospital of Da Lian Medical University from January 2015 to December 2020 and cystic-appearing soft tissue masses.A total of 51 patients were collected,including 28 males and 23 females,ranging in age from 23 to 82 years,with an average age of 47±11.4years.2.MRI equipment and methods Fifty-one patients underwent T₂WI-FS、T₁WI-FS and T₁WI-FS+C.Siemens Verio3.0T,GE Discovery MR 750W 3.0T and GEHDXT 1.5T MR scanners were used as MRI scanning equipment,and corresponding coils were used according to the lesion site.3.Analysis of conventional MR imaging Two diagnostic radiologists independently diagnosed and reviewed the radiographs.The benign and malignant cases were judged and classified,and the signs of specific cases were described.The observation contents included tumor location, shape,size,boundary,signal and enhancement mode.If the results of two diagnostic radiologists are inconsistent after the diagnostic analysis of the nature and signs of the lesion,a third senior physician will join in to make a judgment after a joint discussion.4.Analysis of radiomic4.1.Tumor segmentation and Radiomics Feature Extraction The original DICOM images of 51 patients with cystic-appearing soft tissue mass were imported into the imaging cloud platform Radcloud（Huiying Medical Technology Co.,Ltd）.T₂WI-FS、T₁WI-FS and T₁WI-FS+C sequences were manually outlined layer by layer along the lesion contour to obtain the overall Volume of interest（VOI）of the tumor,and 51 VOI data sets were obtained for each.The VOI obtained was used to calculate the image radiomic eigenvalues.The feature values are divided into feature-based category and filter-based category,in which the former can be divided into first-order feature,shape and size feature and texture feature.The latter includes wavelet transform and Laplacian Gaussian filter.4.2.Radiomics Feature Selection The extracted eigenvalues were selected by removing low variance selection method,univariate feature selection method,minimum absolute selection method and contraction operator algorithm.Among the features extracted from the above three sequences,the four feature values with the largest absolute values of coefficients were selected as the features adopted in the joint construction model of T₂WI-FS,T₁WI-FS and T₁WI-FS+C.4.3.Establish classification models In this study,a random method was adopted,and the proportion of patients was8:2,respectively,as the training set and the verification set.K-nearest neighbor（KNN）and Support Vector Machine（SVM）are used for feature classification.5.Statistical Analyses SPSS 22.0 was used for data analysis.To analyze whether there was a statistical difference between benign and malignant features of conventional MRI,Fisher’s exact test orχ2 test was used.Receiver operating curve（ROC）curve was used to analyze the benign and malignant results interpreted by radiologists to prove the predictive ability of conventional MRI.Area under characteristic curve（AUC）,sensitivity and specificity were used to evaluate the diagnostic effectiveness of the model. ROC curve analysis was used to prove the predictive power of the image omics features,and to calculate the AUC value and the prediction accuracy.Accuracy,recall rate,specificity,and F1-score were used to evaluate the performance of the classifier.Results:1.Routine MRI findings of cystic-appearing soft tissue masses and diagnostic efficacy of them in benign and malignant differentiation The results showed that the homogeneity of T₁WI-FS+C signal and the degree of enhancement were significantly different between the benign and malignant groups of cystic-appearing soft tissue masses（P<0.05）.There were no significant differences in other variables between the malignant group and the benign group（P>0.05）.ROC curve was used to analyze the diagnostic efficacy of conventional MRI in differentiating 51cases of cystic-appearing soft tissue masses.The AUC and P values were 0.762 and 0.001,respectively.The corresponding sensitivity and specificity were 64.0%and88.5%.2.Radiomic diagnostic efficacy of cystic-appearing soft tissue masses in benign and malignant differentiation The combination of T₂WI-FS,T₁WI-FS,and T₁WI-FS+C imaging models ca distinguish benign and malignant cystic and liquid-like soft tissue masses.The diagnostic performance of KNN and SVM is similar.2.1 Diagnostic results and evaluation of KNN classifiers for T₂WI-FS,T₁WI-FS and T₁WI-FS+C sequence joint models:1)The AUC and F1-score of the benign group were 0.87,（accuracy 0.83,sensitivity0.88,specificity 0.88）,and 0.86;2)The AUC and F1-score of the malignant group were 0.87,（accuracy 0.91,sensitivity0.88,specificity 0.88）,and 0.89.2.2 Diagnostic results and evaluation of SVM classifiers for T₂WI-FS,T₁WI-FS and T₁WI-FS+C sequence joint models:1)The AUC and F1-score of the benign group were 0.88,（accuracy 0.76,sensitivity0.79,specificity 0.80）,and 0.80;2)The AUC and F1-score of the malignant group were 0.88,（accuracy 0.83,sensitivity0.80,specificity 0.79）,and 0.79;3.Comparison of diagnostic efficacy between Image doctor and radiomic By De Long test,the ROC difference between the diagnostic efficacy of radiologist and the combined sequence model of radiomics was significant（P=0.017）,and the final AUC value of the radiomics method（0.88）was significantly higher than the AUC value of the radiologists（0.76）.Conclusion: Through the study of MR radiomic with T₁WI-FS,T₂WI-FS and T₁WI-FS+C sequences of 51 cases of soft tissue mass with cystic-appearing,and the comparison of the diagnostic efficacy with that of imaging specialists,the following conclusions can be drawn:1.The radiomic model（KNN,SVM classifier）based on T₁WI-FS,T₂WI-FS,and T₁WI-FS+C sequences can be used to distinguish benign and cystic-appearing malignant soft tissue masses with,and its diagnostic efficacy（AUC）is 0.88.2.Radiomic models have better diagnostic performance than radiologists using the same sequence image reading analysis.The AUC of the former was 0.88,while the AUC of the radiologist was 0.76.