Correlation Between CT Texture Analysis And Gene Mutation In Colorectal Cancer Liver Metastasis

Objective:To explore the relationship between the margins and the tumor itself of colorectal cancer liver metastases and the mutations of KRAS,NRAS,and BRAF genes using CT texture features.Methods: 123 eligible patients were enrolled,and three methods were used(Method 1.Outline the 1cm ring area and the inner area of the lesion;Method 2.Outline the 1cm ring area and the 0.5cm inner area of the lesion;Method 3.The peripheral 1cm ring region)to draw a two-dimensional ROI of the largest cross-sectional area of a single largest liver metastasis on the baseline portal vein CT image.The A.K software was used to extract texture features,reduce dimensions,and build a machine model,construct an optimal machine learning model,obtain the AUC value of the validation set,and analyze and predict performance.Collect clinical features of patients,use SPSS22.0 software to analyze meaningful features and combine them with selected texture features,establish a logistic regression model,obtain ROC curves and AUC values,and compare the performance of the classifier.Results: All three methods used support vector machine(SVM)to build the optimal machine learning model,and the AUC values of the validation set were all greater than 0.7.The AUC values of the validation methods of the three methods were 0.774,0.763,and 0.732,showing good prediction performance.The clinical features of the primary tumor location,the pre-treatment CA199 level,and the number of metastatic lesions were included in the study(P <0.05)and combined with the selected texture features,the AUC values reached 0.79,and the AUC value of method 1 combined diagnosis reached 0.810.All achieved good prediction performance.Conclusion: The imaging texture parameters and clinical features of CRLM patients based on CT portal phase images have better predictive performance in identifying whether KRAS,NRAS,and BRAF genes are mutated.