Click here to


Are you sure ?

Yes, do it No, cancel

Sub-Region Based Radiomics Analysis for Survival Prediction in Esophageal Tumors Treated by Radiotherapy

P Yang1*, L Xu1 , Z Cao1 , Y Jiang1 , Y Xue1 , C Luo1 , S Wu2 ,Y Kuang3 , T Niu1 , (1) Zhejiang University, Hangzhou, Zhejiang, Peoples R China,(2) Hangzhou Cancer Hospital, Hangzhou, Zhejiang, Peoples R China,(3) University of Nevada, Las Vegas, Las Vegas, NV


(Tuesday, 7/16/2019) 1:45 PM - 3:45 PM

Room: Stars at Night Ballroom 2-3

Purpose: Precise prediction of patients' survival and prognosis is clinically important in the treatment of esophageal cancer(EC). Prognostic biomarkers like VEGF, cyclin D1, and Ki-67 are all assessed on the postoperative acquired tumor samples. The existing prognostic approach is difficult to apply to patients who are unable to receive surgery and hinders the preoperatively individualized therapeutic decisions making. Thus, we develop and validate a sub-regional radiomics model for preoperative prediction of overall survival risk in patients with EC.

Methods: A total of 133 EC patients who undergo radiation therapy from two institutions were included in this study. The patients from Institution II (n=46) were used to validate the developed radiomics model. The tumor region on planning CT was automatically clustered into several sub-regions to fully characterize intra-tumor variation. Radiomics features were extracted from each sub-region. The LASSO for Cox regression method was used for feature selection and constructing the overall survival prediction model. The C-Index and receiver operating characteristic(ROC) curves for 3-year survival were employed to assess the prognostic value of the model developed. The patients were stratified into a high-risk, and low-risk group based on predicted risk and Kaplan-Meier analysis was used for survival analysis.

Results: The tumor was partitioned into four distinct sub-regions, and we extracted 548 radiomics features from each sub-region. Seven sub-regional radiomics features were selected to build the overall survival prediction model. The C-indexes of the proposed model were 0.729(0.656-0.801,95%CI) and 0.705(0.628-0.782,95%CI) in the training and validation cohort, respectively. The 3-year survival ROC curve showed an AUC of 0.811(0.670-0.952,95%CI) in the training cohort and 0.805(0.638-0.973,95%CI) in the validation cohort. The Kaplan-Meier analysis showed a significant difference(p<0.001) between the survival of the high-risk and low-risk group.

Conclusion: The proposed sub-regional radiomics model could preoperatively predict the overall survival risk in patients with EC treated by radiotherapy.


CT, Radiation Therapy, Quantitative Imaging


IM- CT: Quantitative imaging/analysis

Contact Email