基于CT影像组学特征预测低分化胆囊癌的价值OACSTPCD

Objective:To explore the value of multiphase CT-based radiomics in predicting poorly differentiated gallbladder carcinoma.Methods:Patients with gallbladder carcinoma confirmed by pathological examinations after radical resection in our hospital were analyzed retrospectively.According to pathological examinations,they were divided into low differentiation(44.5%)and non-low differentiation groups(55.5%),respectively.The region of interest(ROI)was manually delineated by using 3D Slicer software,and the radiomics features of arterial phase and venous phase were extracted respectively and screened by mutual information regression method finally.The training set screened by the three-step dimensionality reduction method was fitted to the K-nearest neighbor algorithm to construct the low-differentiation prediction model of gallbladder carcinoma.The testing set was used to evaluate the model prediction efficiency.The receiver operating characteristic(ROC)curves were drawn,and the area under the curve(AUC)was conducted to determine the predictive power of radiomic features in distinguishing poorly differentiated gallbladder carcinoma.Results:A total of 1 502 dual-phase radiomics features were screened out,and six features were extracted by three-step dimensionality reduction,large area emphasis,large area high gray level emphasis,run entropy,mean value,root mean squared,and 10 percentile.The results of the prediction model,the training set AUC was 0.83(sensitivity 0.76,specificity 0.71),and the testing set AUC was 0.68(sensitivity 0.67,specificity 0.60).Conclusion:The dural-phase CT-based radiomics features have certain value in in predicting the grading of poorly differentiated gallbladder carcinoma and can be repeated well.