Abstract

Info

Title:: Preliminary study of radiomics features of EPID image in detecting setup errors in eye lens radiotherapy

Author(s):: ZHANG Xiangbin1; DAI Guyu1; 2; XIAO Qing1; PENG Xudong1; LI Guangjun1; BAI Sen1; 2; 1. Department of Radiotherapy, West China Hospital of Sichuan University, Chengdu 610041, China 2. West China School of Medicine, Sichuan University, Chengdu 610065, China

Keywords:: Keywords: eyelens radiotherapy electronic portal imaging device transmission dosimetry radiomics setup errors

Abstract:: Abstract: Objective To investigate the feasibility of using the radiomics features of electronic portal imaging device (EPID) image to detect the setup errors in eye lens radiotherapy. Methods A total of 8 patients who underwent eye lens radiotherapy were enrolled in the study. Verification plans were created based on head phantom to simulate treatment delivery. Taking the original EPID image which was obtained without setup errors as the reference, the EPID images of different setup errors were generated. The radiomics features of the EPID images were extracted with Pyradiomics tool, and whether radiomics features could be used for setup errors detection in clinic was evaluated. Results Seventy-four out of 107 selected features were observed significant correlation with the vector length of 3 directions with setup errors (P<0.05). The clinical setup errors detection based on radiomics features was superior to that based on traditional gamma analysis. The area under curve of traditional gamma analysis was only 0.79, while that of the most relevant univariate feature extracted by radiomics reached 0.84, and that of multivariate feature based on Ridge regression was 0.90. Conclusion For the detection of setup errors based on EPID image, traditional gamma analysis has some limitations, and it has been demonstrated that radiomics features-based method has better prediction performance and greater application potential.