Objective To investigate the efficacy of a combined PET/CT radiomics model in predicting liver metastasis in patients with colorectal cancer.Methods A retrospective analysis was conducted on 73 patients with primary colorectal cancer treated at Shaoxing Central Hospital.Based on postoperative pathology results or MR enhancement tests,patients were divided into a liver metastasis group(28 cases)and a non-liver metastasis group(45 cases).Lesions were manually delineated in PET/CT images,from which radiomics features were extracted after preprocessing the area of interest.Prelim-inary selection of stable and reliable features was conducted using ICC analysis,analysis of univariate and variance.Then,CT and PET radiomics features were progressively selected on the training set by combining different feature selection meth-ods(mRMR and Lasso).Based on the filtered features,multiple single-modality radiomics signatures were established u-sing three machine learning algorithms.Lastly,the LR method was used to construct a combined model of the two optimal single-modality radiomics labels,to explore the most effective combined PET/CT radiomics model for colorectal cancer liver metastasis.Results In the single-modality radiomics models,10 radiomics features were respectively included in the CT and PET-related models for modeling.The optimal CT radiomics model based on LR(MLR_CT)achieved an AUC,sensi-tivity,and specificity of 0.701,66.7％,and 58.3％,respectively,on the test set.The optimal PET radiomics model based on the SVM(MSVM_PET)reached an AUC,sensitivity,and specificity of 0.811,72.7％,and 91.7％,respec-tively,on the test set.The combined PET/CT radiomics model(MCT+PET)attained an accuracy of 100％for all metrics on the training set,and high AUC,sensitivity,and specificity of 0.92,81.8％,and91.7％respectively,on the test set.Conclusion The combined PET/CT radiomics model significantly enhances the efficacy in predicting liver metastasis in colorectal cancer patients compared to single imaging modalities,holding great potential as an auxiliary tool to improve the accuracy of clinical diagnosis of colorectal cancer liver metastasis.
维普
万方数据
