| Objective:Different subtypes have different clinical biological behaviors,response to treatment and prognosis of patients.Judging the nature of tumors has always been one of the concerns and difficulties in clinical work.Accurate preoperative judgment of tumor properties will guide clinical treatment strategies.However,the lack of specificity of conventional Magnetic Resonance Imaging(MR)may lead to inaccurate diagnosis of extremity soft tissue tumors,especially some intermediate tumors.Therefore,there is a clinical need for means that can accurately distinguish the nature of soft tissue tumors in the extremities before surgery.There is a lack of unified understanding of the evaluation of postoperative recurrence of soft tissue sarcoma of the extremities at present.Differences in the assessment of the possibility of recurrence will directly lead clinicians to different therapy,which have an important impact on the quality of life of patients.Therefore,there is a need for a variety of methods that can assess the possibility of postoperative recurrence and provide more comprehensive information for precise treatment.Lipoma is the most common benign fat cell tumor of the extremities.Lipoma and atypical lipomatous tumor/well-differentiated liposarcoma have different biological behaviors and treatment methods,but the similar histological components and imaging features make the two diagnosis be one of the difficulties in clinical work.Whether the difference in fat content between the two is related to histological types has not yet been reported.The appearance of MR fat quantitative sequence makes it possible to quantitatively study fat content,which will help us to have a deeper understanding of tumor biological behavior.With the development of imaging technology,radiomics,as an emerging image analysis technology,achieves accurate judgment and prediction of diseases through highthroughput information acquisition,so it has received more and more clinical attention.However,there are still few studies on radiomics in the evaluation of the properties and prognosis of extremity soft tissue tumors.Radiomics models were built to distinguish benign,intermediate and malignant soft tissue tumors of the extremities,and to explore the feasibility and efficacy of the radiomics model in the first part.In the second part,a radiomics model was constructed to predict the possibility of recurrence of soft tissue sarcoma within two years after surgery,and the recurrence risk of patients was stratified.Cox model was built to quantify the time and recurrence probability and individualized follow-up time and strategy for patients.In the third part,we aim to explore the differences in the proportion of fat in different type of adipocyte tumors by IDEAL-IQ and to differentiate lipomas from well-differentiated liposarcoma/atypical lipomatous tumors.To explore the correlation between FF value and the expression of KI-67,and then to explore the correlation between the difference in tumor fat content and the degree of tumor malignancy;radiomics model based on IDEALIQ sequence was built to predict tumor’s KI-67 expression.Methods:Retrospectively collected a total of 285 cases of soft tissue tumors of the extremities from 2015 to 2019.Image segmentation,feature extraction and selection were based on T1WI,FS T2WI and T1&T2WI combined sequences.Three classifiers were selected to build three-class radiomics models of different sequences.ROC curve,accuracy,precision,Recall and F1-score metrics evaluateed the performance of the model.Delong test was used to evaluate the performance differences of models.Using patient clinical information and MR image characteristicsto establish a clinical model.The performance of the radiomics model was compared with the clinical model to assess the ability of the radiomics model to identify the nature of soft tissue tumors.Retrospectively collected patients with soft tissue sarcoma tumors of the extremities from 2015 to 2019.Five classifiers were selected to build models for predicting tumor recurrence.The radiomics model with the best diagnostic performance was selected,and the survival analysis was performed according to the best cut-off value of its RADSCORE.The recurrence risk within 2 years after the operation was stratified according to the survival curve.Nomogram for predicting the risk of recurrence was built by Cox model.Harrell Concordance Index(C-Index)and the calibration curve to assess the power of the nomogram.Net Reclassification Index(NRI)and Decision Curve Analysis(DCA)were used to evaluate the clinical application value of nomogram.Patients with extremity adipocyte tumors from 2018 to 2021 were collected,preoperative IDEAL-IQ sequence were scanned.One-way analysis of variance It is used to compare the difference of FF value between benign,intermediate and malignant tumor groups.Find the best cut-off value according to ROC to distinguish lipoma and welldifferentiated liposarcoma.In addition,benign and non-benign adipocyte tumors were distinguished according to the best cut-off value of the ROC curve.Pearson correlation analysis was used to evaluate the correlation between FF value and KI-67 expression;Radiomic model was used to predict KI-67 expression.Results:In the identification of benign,intermediate and malignant soft tissue tumors of the extremities,the diagnostic performance of the radiomics model constructed by the Support Vector Machine(SVM)classifier is better than other classifiers;the SVM radiomics model based on T1WI,FS T2WI and T1&T2WI combined sequence AUC in the test set were 0.85,0.81 and 0.85,and the accuracy was 0.64,0.61 and 0.72,respectively,and the precision were 0.71,0.63,and 0.64,respectively,Recall was 0.71,0.61,and 0.64,and F1-score was 0.71,0.61,and 0.64,respectively.The AUC of the clinical model was 0.8 and P<0.05 by Chi-Square test.In the identification of intermediate tumors,the AUC of the radiomics model can reach 0.8,demonstrating the ability of the radiomics model to identify the properties of tumors that are difficult to distinguish clinically.The ability of clinical models to identify intermediate tumors was weak,with an AUC of only 0.74,which was lower than the performance of radiomics models.The radiomics model constructed by the SVM classifier using T1&T2WI combined sequences to predict the 2year postoperative recurrence of soft tissue sarcoma of the extremities had the best performance in 5-fold cross-validation(testing set AUC0.92;accuracy 0.87,sensitivity 0.89,specificity sex 0.85).The training set and test set were divided into high risk group(RADSCORE>0.47)and low risk group(RADSCORE<0.47)using the best cutoff value of 0.47.KI-67 and RADSCORE became independent predictors.The combined Cox hazard proportional regression model constructed based on these two variables had better performance in predicting the recurrence.C-Index was 0.879 and 0.851 in the training set and the test set respectively.Nomogram was constructed to predict the recurrence time and recurrence probability within 2 years after sarcoma surgery.The clinical efficacy of the combined model was excellent,and the clinical benefit of the combined model was 16.7%and 62.5%higher than that of the single RADSCORE model and the single KI-67 model,respectively;in addition,the single RADSCORE model was 54.2%higher than the KI-67 model.ANOVA showed statistically significant differences in FF values between different adipocyte tumor groups;According to the ROC,the best cut-off value 84.9%can distinguishing lipoma from well-differentiated liposarcoma/atypical lipomatous.The AUC was 0.87,the accuracy was 0.86,the specificity was 0.9,and the sensitivity was 0.8.At the same time,84.9%is also the best cut off value for distinguishing benign from nonbenign adipocyte tumors.The AUC of the ROC curve at this optimal cut-off value is 0.94,the accuracy is 0.91,the sensitivity is 0.91,and the specificity is 0.9.FF value was significantly negatively correlated with the expression of KI-67;the regression model constructed based on the radiomic features of IDEAL-IQ sequence can predict the expression of KI-67 in adipocyte tumors.In the testing set,Explained Variance Score(EVS)is 0.91,Mean Absolute Error(MAE)is 1.92,Mean Square Error(MSE)is 14.21;R2 score is 0.9.Conclusion:SVM models constructed based on T1WI,FS T2WI and combined sequences had similar performances;the diagnostic performance of the models constructed by radiomics was higher than that of the clinical models.Therefore,the radiomics model can better assist doctors in making a diagnosis in clinical practice,especially for the diagnosis of intermediate tumors,which will assist clinicians in comprehensively evaluating the nature of tumors and provide a basis for patients to correctly formulate treatment plans.The radiomics model constructed based on the combined sequence of T1&T2WI to predict the recurrence of sarcoma patients within 2 years after surgery has high performance,and the recurrence risk stratification analysis was successfully achieved by using the best cut-off value of RADSCORE;the combined nomogram can be used for sarcoma recurrence time and recurrence Probability to predict;the combined model has a large clinical net benefit.The probability of 2-year recurrence time of sarcoma can be predicted by nomogram,and individualized follow-up time and follow-up strategy are formulated for patients.Patients with high recurrence risk can be monitored after surgery accordingly.Radiomics models and nomogram information provide more information for personalized precision medicine.In this study,MR fat quantitative sequence was used to quantitatively describe the fat content in adipocyte tumors.The proportion of fat content in adipocyte tumors was different,and the difference in fat content could reflect the difference in tumor tissue composition.We used the best cut off of ROC curve to distinguish different tumors for the first time.This method can provide more diversified information for clinically distinguishing these two types of tumors;it also differentiated benign and non-benign adipocyte tumors which might lead different clinical treatment method.In addition,the FF value was significantly negatively correlated with the expression of KI-67,indicating that the difference in intratumoral fat content may reflect the difference in the degree of tumor cell proliferation.We also used MR quantitative images and radiomics to predict the expression level of KI-67 in extremity adipocyte tumors for the first time.The results of this study showed that the radiomics model can predict the expression level of tumor KI-67.The combined study of quantitative imaging and KI-67 expression provided more information for a comprehensive evaluation of the biological behavior of lesions. |
PDF Full Text Request