| Purpose:To predict microsatellite instability(MSI)status of gastric cancer on preoperative dual-energy CT imaging using quantitative parameters and radiomics analysis.Materials and Methods:123 patients with gastric cancer who were pathologically confirmed were involved in this study.They underwent preoperative revolution CT enhanced spectral imaging scan.According to the state of MSI tested by immunohistochemical examination,the patients were divided into MSI(41 patients)and MSS(82 patients)groups.We collected the clinicopathological date of the patients,including age,gender,tumor size,tumor gross type,WHO histology type,histological differentiation degree,TNM stage,tumor infiltration depth,lymph node metastasis,distant metastasis,neurovascular invasion.In the first part of the study about“Value of dual-energy CT quantitative analysis in the prediction of MSI status of gastric cancer”,Two radiologists used a double-blind method and placed three regions of interest(ROIs)on lesions of the slice covering the largest dimension of tumor to measure the CT values of 40-70 ke V virtual monochromatic images(VMIs)on pre-contrast and enhanced phases,the effective atomic number(Eff-Z)during the pre-contrast CT scan,iodine concentration(IC)and normalized iodine concentration(NIC)on enhanced phases.The average values of three ROIs were calculated to minimize measurement bias.Interobserver reliability is described with the intraclass correlation coefficient(ICC).The comparison of each parameter between the MSI and MSS groups was analyzed with the independent sample t test or Mann-Whitney U test.Logistic regression analyses were used to identify independent factors and combination diagnosis.Receiver operating characteristic(ROC)curves were generated,and the Delong test was performed to evaluate and compare the diagnostic performance of each parameter and their combination.In the second part of the study about“Radiomics analysis of multi-energy images with dual-energy CT for predicting MSI status in gastric cancer”,we reconstructed 40ke V VMIs on pre-contrast CT(PC)and 50 ke V VMIs on venous phase(VP)and imported them into AK software for image preprocessing.Two readers manually delineated the lesion contour on the largest axial diameter of the gastric cancer and its adjacent upper and lower slices using the ITK-SNAP software.107 radiomics features were extracted from the tumor region on each phase image using AK software.All patients(n=123)were randomly divided at a ratio of 7:3 into training(n=86,MSS=57,MSI=29)and validation cohorts(n=37,MSS=25,MSI=12).The consistency test,Spearman correlation test and gradient boosting decision tree(GBDT)were used for select the representative features.The selected features and Rad-scores were used to develop radiomics models for MSI status on the basis of multivariable logistic regression by fivefold cross-validation.The radiomics models included:(1)model based on PC 40 ke V VMIs;(2)model based on VP 50 ke V VMIs;(3)combined model based on PC 40 ke V and VP 50 ke V VMIs.Receiver operating characteristic curve(ROC),decision curve analysis(DCA),and calibration curve(Hosmer-Lemeshow test)were used to evaluate each radiomics model.We used univariate and multivariate logistic regression analyses to identify the independent factors for predicting MSI status in gastric cancer among clinicopathological characteristics and establish clinical model.We further combined the clinicopathological model and the optimal radiomics model to establish clinicopathological-radiological model and presented the combined model as nomogram.Then,to determine the optimal MSI prediction model,we evaluated the clinicopathological model,the radiomics model and combined model using the same method as before.Results:In the first part of the study,the consistency of the data obtained by the two observers was good(ICC value>0.75).The CT values of 40-60 ke V and Eff-Z values on pre-contrast CT in the MSI group and MSS group were 67.24(63.33,79.16)vs.63.70(58.27,70.19)HU,54.86(51.72,60.97)vs.52.39(47.30,57.49)HU,47.99±5.50 vs.45.10±8.65 HU,7.86(7.82,7.94)vs.7.85(7.78,7.90),respectively(P<0.05);the corresponding AUC values of these parameters in predicting MSI status were 0.630,0.626,0.616,0.615,respectively.The CT values of 40-70 ke V and IC values on arterial phase were 159.88(142.23,180.59)vs.178.10(151.10,212.50)HU,115.66(103.48,130.21)vs.128.59(111.46,152.49)HU,89.15(80.51,99.29)vs.98.12(87.88,114.34)HU,73.30±13.56 vs.81.26±18.44 HU,15.68(13.34,17.77)vs.17.83(14.10,21.28)100ug/cm~3,respectively(P<0.05);the corresponding AUC values were 0.632,0.636,0.637,0.637,0.622,respectively.The CT values of 40-70 ke V,IC and NIC values on venous phase were 194.85(183.14,222.54)vs.239.72(205.55,282.36)HU,143.14(131.45,159.37)vs.169.35(146.27,199.39)HU,108.01(100.16,118.62)vs.127.35(111.35,148.34)HU,85.86(79.81,94.30)vs.99.99(88.58,113.92)HU,19.75(18.14,22.69)vs.24.76(20.87,29.39)100ug/cm~3,0.10(0.08,0.12)vs.0.11(0.09,0.14),respectively(P<0.05);the corresponding AUC values were 0.719,0.723,0.718,0.713,0.711,0.670,respectively.The CT values of 40-70 ke V,IC and NIC values on delayed phase were 195.09±34.65 vs.224.21±56.29 HU,139.67±23.51 vs.159.24±37.08 HU,105.58±16.88 vs.119.17±25.57 HU,84.35±12.99 vs.94.30±18.65 HU,19.20(16.85,21.54)vs.22.16(18.25,26.77)100ug/cm~3,0.48(0.41,0.54)vs.0.41(0.36,0.49),respectively(P<0.05);the corresponding AUC values were 0.680,0.685,0.690,0.693,0.663,0.685,respectively.The logistic regression analysis showed the independent factors for predicting MSI were CT values of PC 40 ke V and VP 50 ke V VMIs.The combined diagnostic performance of the two independent factors(AUC=0.746)was higher than that of each individual parameters,where the differences between AUC(PC40 ke V+VP 50 ke V)and AUCs of all parameters on pre-contrast CT and IC values on arterial phase were statistically significant(P<0.05).In the second part of the study,The AUCs of the combined radiomics model based on PC 40 ke V and VP 50 ke V VMIs in the training and validation cohorts were 0.808,0.810,respectively,which showed higher performance than any other radiomics model,but the differences between AUCs of these models were not significant(P>0.05).The combined radiomics model showed larger clinical application value than PC 40 ke V and VP 50 ke V models.All radiomics models showed good calibration effects.Logistic regression analysis showed the independent clinicopathological factors for predicting MSI were tumor location and TNM stage.The AUCs of the clinicopathological model in the training and validation cohorts were 0.714 and 0.717,respectively.We further combined the clinicopathological model and the optimal radiomics model(i.e.,combined radiomics model based on PC 40 ke V and VP 50 ke V VMIs)to establish clinicopathological-radiological model,with AUCs of 0.870,0.833 in the training and validation cohorts,respectively.In the training cohort,the combined model were significantly higher than the clinicopathological model(P=0.003).In the test cohort,the differences between AUCs of the three models were not significant(P>0.05).The combined model showed larger clinical application value than the clinicopathological model and radiomics model,and all models showed good calibration effects.Conclusions:1.The CT values of 40-60 ke V and Eff-Z values on pre-contrast CT,CT values of40-70 ke V and IC values on enhanced phases,and NIC values on venous and delay phases can effectively differentiate MSI from MSS gastric cancer.The CT values of PC40 ke V and VP 50 ke V VMIs were independent factors for predicting MSI status of gastric cancer,and the combination of the two parameters can improve the differential diagnosis efficiency.2.The radiomics model based on multi-energy images with dual-energy CT has a certain value in predicting MSI status of gastric cancer.The model combined clinicopathological and radiomic features achieved the best predictive performance,which has the potential to be used in routine clinical practice as new indicator to evaluate the MSI status of gastric cancer. |
PDF Full Text Request