The Impact of the Region of Interest on the Prediction of the Degree of ICC Differentiation in MRI Radiomics Models
Objective To investigate the effect of different ranges of interest(ROI)on MRI radiomics models to predict the degree of pathological differentiation of intrahepatic cholangiocarcinoma(ICC).Methods A total of 191 patients with ICC confirmed by postoperative pathology were retrospectively collected and randomly divided into a training group(n=133)and a validation group(n=58)according to a 7:3 ratio.Two physicians manually sketched ROIs layer by layer along the tumor edge on MRI scanning and enhancement sequences,followed by automatic outreach of 8,10 and 12 mm by software and manual adjustment to obtain ROItumor,ROI8mm,ROI10mm and ROI12mm.The radiomics features of the above four ROIs were extracted respectively,and the features were screened by the least absolute shrinkage and selection operator(LASSO),the model was built using logistic regression and validated with the validation group.Finally,the area under the characteristic curve(AUC),calibration curve and decision curve were used to evaluate the model performance.Results The highest AUC of the 10mm model in the validation group was 0.754,and the calibration curves of all models performed well in the Hosmer-Lemeshow test(P＞0.05).Decision curves showed the highest benefit was for the 10mm model when the threshold was 0.24.Conclusion The choice of ROI range has an impact on the degree of differentiation of ICC predicted by MRI radiomics models,and 10 mm may be the optimal range.
NETL
NSTL
万方数据
维普
