Assess The Value Of Radiomics In Predicting The Recurrence And Metastasis Of T3 Rectal Cancer

Objective:To assess the value to predict recurrence and metastasis of T3 rectal cancer by radiomics.Method:A retrospective study of 87 patients with T3 rectal cancer at the First Hospital of Jilin University from January 2011 to December 2013.All the patients underwent contrast-enhanced abdominal CT within 2weeks before surgery,including arterial phase and vein phase.Clinical factors,such as gender,age,bloody stool,tumor differentiation,lymph node metastasis,carcinoembryonic antigen(CEA),carbohydrate antigen 72-4(CA72-4),carbohydrate antigen 199(CA199)and CT datasets were obtained.The radiology group cloud platform(http://www.radcloud.cn)was used to map ROI of the lesion and 1029 radiographic features were extracted.All patients were assigned to two groups:training set(n=44)and validation set(n=43).Using training set data for univariate analysis,select radiomics features associated with recurrence and metastasis,significantly.Then establishment a predictive model for recurrence and metastasis by Cox analysis.Apply validation set data to verify the predictive model by Kaplan-Meier curves and evaluate the difference between clinical predictive model and final predictive model by Log-rank test.Results:There was no significant difference in clinical factors between the training set and the validation set.Seven arterial phase radiomics features and nine venous phase radiomics features predict the recurrence and metastasis of T3 rectal cancer after surgery strongly(CI > 0.60 and FDR < 5%).The radiomics features that were finally selected into the predictive model were wavelet HHHGLRLM-short run high gray level emphasis,and the arterial phase and venous phase had the same results.Among the clinical factors,histological grade and lymph node metastasis were included in the predictive model(Log-Rank test,p < 0.1).Compared with the CI index of the predictive model using clinical factors alone,the CI index of the final predictive model of radiomics features of arterial phase increased from 0.69 to 0.76,and the CI index of the final predictive model of radiomics features of venous phase increased from 0.69 to 0.74.Conclusion:For predicting recurrence and metastasis of T3 rectal cancer,the addition of radiomics features can improve the prognostic ability of a model compared to one using clinical factors alone.