| Objective: To construct a joint pattern for predicting microsatellite instability high(MSIH)status in patients with endometrial endometrioid carcinoma(EEC)based on magnetic resonance imaging(MRI)characteristics and radiomic features.Methods : In the research,imaging and clinical medical data of 144 patients with EEC have been pathologically confirmed are retrospectively collected in Liaoning Cancer Hospital from 2018-2021.According to the immunohistochemical results(MLH1,MSH2,MSH6 and PMS2)reported by pathology,patients were divided into MSI-H group(n=48)and non-MSI-H group(n=96).Clinical characteristics(age,menopause status,body mass index(BMI),blood routine indicators and tumor markers)and conventional MRI diagnostic characteristics(degree of myometrial invasion、maximum diameter of tumor、cervical invasion and lymph nodes status)were analyzed.The Clinical risk opportunities of MSI-H were extractedly based on the clinical conventional MRI diagnostic features,and the routine MRI characteristic model(M1)was developed using logistic regression analysis.The two radiologists used ITK-SNAP software(version 3.6.0)to segment the region of interest(ROI)of MRI images(axial T2 W and axial contrast-enhanced T1W).Import annotated image into AK software(version 3.2.0)and set three major categories of texture parameters to extract radiomic signature.Calculate intraclass correlation coefficient(ICC)by using IPMs(version 2.5.0)to evaluate consistency in segmentation of lesions between two radiologists.ICC not less than 0.75 are considered stably.A total of 144 patients were randomly divided into a training group(n=100)and a validation group(n=44)in a 7:3 ratio.Dimensionality reduction and screening of radiomics signature set were performed by Correlation＿XX、GBDT analysis and Multivariate-Logistic,calculate the radiomics score(radscore)and construct the radiomics model(M2).A joint model(M3)was constructed using logistic regression analysis combined with radiomic signature and routine MRI characteristics and draw a nomogram.R software(version 4.2.2)was used for receiver operating characteristic(ROC)curve analysis to assess predictive efficacy,and the area under the curve(AUC),specificity,and sensitivity were computed.De Long test was performed for testing the difference among AUC values.In order to assess the calibration performance,potential clinical value and improvement of classification efficacy of models.Calibration curves and decision curve analysis of the individual models were plotted.Net reclassification index(NRI)and integrated discriminant improvement index(IDI)were calculated.Results:1.Between the training and validation groups,the differences among patients in the non-MSI-H and MSI-H groups were not statistically significant(p>0.05),except for the differences in the depth of myometrial infiltration and the interstitial condition of the cervix,which were statistically significant(p<0.05).The routine MRI characteristic model(M1)was obtained by logistic regression analysis,which was composed of the judgment of the invasion of cervical stroma and the depth of muscular invasion on MRI,AUC =0.720 in the training group and AUC=0.692 in the validation group.2.A total of 2632 radiomics features were extracted from two sequences of images,horizontal axial T2 W and horizontal axial contrast enhancement T1 W,and finally 6radiomics features(ICC= 0.845-0.991)building models(M2)were selected to predict MSI-H status.AUC in training group was 0.843,and AUC in validation group was 0.759.3.A combined model(M3)(consist of a routine MRI characteristic model and a radiomics signature model)was constructed with AUC=0.898 for the training group and AUC=0.823 for the validation group.Calibration and decision curve analysis indicated that the model performed best and had the highest clinical application value.4.The diagnostic efficacies of M3 versus M1 differed in both the training and validation groups,and the differences were statistically significant(p<0.001,p=0.017,respectively).The diagnostic efficacy of M2 versus M1,differed significantly in the training group(p=0.033),but not in the validation group(p=0.600).The diagnostic efficacy of M3 versus M2 differed significantly in the training group(p=0.013),but not in the validation group(p=0.421).5.In the training group and validation group,M3 showed positive classification improvement compared with M1 and M2(NRI>0,IDI>0,p<0.05).Conclusion: The radiomics model based on the radiomics signature of MRI images can be a good predictor of EEC MSI-H status.The combined model of conventional MRI feature model and radiomics model can effectively improve the prediction ability and provide important reference value for clinical understanding of MSI-H status... |
