Assess The Value Of Radiomics For Lymph Node Metastasis In Colorectal Cancer

Objective: To explore the diagnostic value of CT-based radiomics for lymph node metastasis in colorectal cancer.Method: In this retrospective study,446 patients with pathologically confirmed by surgery to colorectal cancer and preoperative contrast-enhanced CT examination in the first hospital of Jilin University from December 2016 to December 2017 were included(212 patients with lymph node metastasis and 234 patients without lymph node metastasis).The patients were divided into training dataset(356 patients)and verification dataset(90 patients)by computer random software according to the ratio of 4:1.All portal venous-phase CT images were retrieved from PACS for image segmentation and analysis.ROI1 was the tumor legion on the maximal diameter axial image slice.ROI2 was the tumor legion on the maximal diameter reformatted image slice.Each region of interest was extracted 1209 radiomics features respectively.13 radiomics features were selected by lasso process for ROI1,11 radiomics features for ROI2.The Radiomics features selected were constructed as the corresponding radiomics signature.Multivariable random forest classifier(RFC)model was chose to establish radiomics model and combined model with clinical information.The postoperative pathological results were used as the gold standard to evaluate the diagnostic efficacy.Results: There was no significant difference in clinical factors between the training dataset and the verification dataset.The radiomics signature,age,sex,location,carcinoembryonic antigen(CEA),carbohydrate antigen19-9(CA19-9)expression were showed correlation with the lymph node metastasis of colorectal cancer(P<0.05).As for the classifcation performance of the radiomics RFC in lymph node metastasis,the AUC of training dataset were 0.998 and 0.997,in the verification dataset obtained the AUC of 0.680 and 0.713,respectively 95%CI: 0.567 to 0.793?0.605 to 0.820.As for the classifcation performance of the RFC combining the radiomics signature and clinical characteristic,the AUC of training dataset were 0.954 and 0.941.in the verification dataset,the AUC were 0.803 and 0.809,respectively 95%CI: 0.709 to 0.897?0.722 to 0.896.Conclusion: The RFC model integrated with the radiomics signature based on CT imaging and clinical characteristic could be used for diagnosing lymph node metastasis in colorectal cancer.