Radiomics Model Based On IVIM-MRI And DKI-MRI For Predicting Histopathological Grade And Ki-67 Expression Level Of Soft Tissue Sarcomas

Purpose:To explore the feasibility and value of predicting the histopathological grade of soft tissue sarcomas（STSs）and Ki-67 expression level by Radiomics model based on IVIM-MRI and DKI-MRI parameter maps（Standard ADC,D,D*,f,MK,MD）Materials and Methods:1.Clinical data42 patients diagnosed with STSs pathologically from May 2018 to January 2020from the Second Affiliated Hospital of Dalian Medical University were collected,including 23 males and 19 females,aged from 12 to 85 years old with an average age of48.5 years old.2.MRI equipment and methodGE Discovery MR 750W 3.0T superconducting magnetic resonance scanner was used for examination.Knee joint coil,large flexible coil,head-neck joint coil and16-channel phased array coil were used to perform conventional MRI scan,IVIM-MRI and DKI-MRI on all patients.3.Image post-processing（1）.Calculation of IVIM-MRI and DKI-MRI parameter mapsUse the MADC software in Functool of GE ADW 4.7 workstation to process the IVIM and DKI images of 42 STSs to obtain Standard ADC,D,D*,f,MD and MK grayscale diagrams.（2）.STSs lesion segmentationThe ITK-SNAP 3.8.0 software was used to outline the Volume of Interest（VOI）on Standard ADC,D,D*,f,MD and MK grayscale maps of 42 cases of STSs layer by layer,covering all areas of the tumor.（3）.Extraction of Radiomics featuresUpload all VOI files and parameter maps to the Radiomics cloud platform（http://mics.radcloud.com）of Huiying Medical Technology（Beijing）Co.,Ltd for Radiomics feature value extraction.A total 1409 Radiomics features of two categories（based on feature classes and Filters）were extracted.（4）.STSs histopathological classification determination and Ki-67 expression level detection（1）Judgment of histopathological classificationAccording to FNCLCC’s STSs histopathological classification standard,42pathological specimens of STSs were tissue sectioned by pathologists,and tumor solid components were selected to observe the degree of differentiation of tumor cells,necrotic area of tumors and mitotic images for giving FNCLCC scores.Grade I is defined as low grade,grade II and III are defined as high grade.（2）Detection of Ki-67 expression levelWith the assistance of pathologists and physicians,routine HE stains and immunohistochemical detection of Ki-67 expression levels were performed on 42 STSs pathological specimens.STSs patients were divided into Ki-67 high expression groups（Ki-67≥25%）and Low expression group（Ki-67<25%）.（5）.Establishment of Radiomics model（1）STSs IVIM-MRI and DKI-MRI parameter map dimensionality reductionVariance Threshold,Select KBest,and Least absolute shrinkage and selection operator（LASSO）were used for three-dimensional dimensionality reduction analysis.（2）STSs IVIM-MRI and DKI-MRI parameter map machine learningThe Radiomics features of 42 STSs IVIM-MRI and DKI-MRI parameter maps after three-dimensional dimensionality reduction were used as the data set,of which 80%was used as the training set to train the machine learning model,and the remaining 20%was used as the validation set to evaluate the accuracy of the model.The area under the ROC curve,95%confidence interval,sensitivity,specificity and F1-Score of the six classifiers（Random Forests（RF）,K-Nearest Neighbor（KNN）,Support Vector Machine（SVM）,Extreme Gradient Boosting（XGBoost）,Logistic Regression（LR）and Decision Tree（Decision Tree,DT））of training set and validation set of the IVIM-MRI and DKI-MRI parameter maps were calculated.Results:（1）.MRI performance of STSs（1）Conventional MRI performancea.Location:25 cases of limbs,10 cases of trunks,3 cases of hips,2 cases of retroperitoneum,and 2 cases of heads;b.Size:The maximum diameter of the tumor was<5cm in 17 cases,and≥5cm in25 casesc.Morphology:26 cases with round shape,16 with irregular shape;d.Boundary:35 cases were clear;7 cases were blurry;e.Homogeneity:11 cases were homogeneous,and 31 cases were heterogeneity;f.Signals:T₁WI showed equal,slightly lower signal in 26 cases and slightly higher signal in 16 cases;T₂WI showed high signal in 11 cases and mixed signal in 31 cases.（2）IVIM-MRI and DKI-MRI performanceOf all 42 STSs IVIM-MRI and DKI-MRI sequences,35 showed high signal and 7showed equal and slightly higher signal.The standard ADC,D,D*,f,MK and MD parameters of 42 STSs showed high and low mixed signal.（2）.Histopathological types,grades and Ki-67 expression level of STSs（1）Histopathological types:16 cases of liposarcoma,3 cases of malignant peripheral schwannoma,3 cases of synovial sarcoma,6 cases of leiomyosarcoma,8cases of fibrosarcoma,1 case of extraosseous chondrosarcoma,3 cases of undifferentiated pleomorphic sarcoma,1 case of malignant giant cell tumor of tendon sheath and 1 case of angiosarcoma.（2）Histopathological grades:15 cases of low grade,27 cases of high grade.（3）Ki-67 expression level:24 cases of Ki-67 low expression and 18 cases of Ki-67high expression.（3）.Results of predicting histopathological classification of STSs based on IVIM-MRI and DKI-MRI parameter maps Radiomics modelA total of 42 patients（48.5±36.5 years old）were included in the study,including15 low grades and 27 high grades.D-SVM achieved the best diagnostic performance.After three steps of dimension reduction,a total of 6 Radiomics features were selected for machine learning.The area under the ROC curve of the validation set is 0.88,the95%confidence interval is 0.53-1.00,the sensitivity is 0.75（low grades）,0.83（high grades）,and the specificity is 0.83（low grades）,0.75（high grades）,F1-Score is 0.75（low grades）,0.83（high grades）.（4）.Results of predicting Ki-67 expression level of STSs based on IVIM-MRI and DKI-MRI parameter maps Radiomics modelA total of 42 patients（48.5±36.5 years old）were included in the study,including24 with Ki-67 low expression and 18 with Ki-67 high expression.MK-SVM achieved the best diagnostic performance.After three steps of dimension reduction,a total of 6Radiomics features were selected for machine learning.The area under the ROC curve of the validation set is 0.83,the 95%confidence interval is 0.50-1.00,the sensitivity is0.83（Ki-67 low expression）,0.50（Ki-67 high expression）,and the specificity is 0.50（Ki-67 low expression）,0.83（Ki-67 high expression）,F1-Score is 0.77（Ki-67 low expression）,0.57（Ki-67 high expression）.Conclusion:By Radiomics study of 42 cases of STSs’IVIM-MRI,DKI-MRI parameter maps（Standard ADC,D,D*,f,MK,MD）with their histopathological grade and Ki-67expression level,we can draw the following conclusions:1.Radiomics classifier based on IVIM-MRI Standard ADC and D parameter maps can identify high-and low-grade STSs,of which D-SVM has the best diagnostic performance in predicting the pathological grade of STSs compare with IVIM-Standard ADC.2.Radiomics classifier based on DKI-MRI MD,MK parameter maps cannot predict the histopathological grade of STSs.3.Radiomics classifier based on IVIM-MRI Standard ADC and D*parameter maps can identify the high and low Ki-67 expression levels of STSs,of which D*-SVM has the best diagnostic performance in predicting Ki-67 expression level of STSs compare with IVIM-Standard ADC.4.Radiomics classifier based on DKI-MRI MK parameter maps can identify high and low Ki-67 expression level of STSs,of which MK-SVM has the best diagnostic performance in predicting Ki-67 expression levels of STSs compare with DKI-MD.