Abstract

Info

Title:: Radiomics combined with interpretable machine learning in predicting the response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer

Author(s):: LI Jianfeng1; SUN Meijuan1; PENG Haiyan2; HU Wenyou2; JIN Fu2; LI Zhaoxia3; WANG Ning3; 1. Postgraduate Training Base in PLA Rocket Force Characteristic Medical Center, Jinzhou Medical University, Beijing 100088, China;2. Radiation Physics Center, Chongqing University Cancer Hospital, Chongqing 400030, China; 3. Department of Oncology, PLARocket Force Characteristic Medical Center, Beijing 100088, China

Keywords:: locally advanced rectal cancer; neoadjuvant chemoradiotherapy; radiomics; machine learning; interpretability

Abstract:: The efficacy of preoperative neoadjuvant chemoradiotherapy (nCRT) in locally advanced rectal cancer (LARC) is predicted using radiomic features of the target areas in radiotherapy for rectal cancer and an interpretable machine learning model. The clinical data are collected from 290 LARC patients who are divided into effective and ineffective groups based on tumor regression grade. The extracted radiomic features and clinicopathological data are used to develop prediction models. The optimal model is determined based on AUC performance evaluation, and the explanatory analysis is conducted using nomogram and decision curve. A total of 223 patients are included in the study, with 48 in the effective group. There are 156 patients in the training set (34 in the effective group) and 67 patients in the validation set (14 in the effective group). The nomogram model shows the best performance, with AUC of 0.858 in the training set and 0.844 in internal test set, and decision curve analysis demonstrated its superior net clinical benefit across most threshold ranges than other models. Combining radiomics and clinical variables, the nomogram can effectively predict nCRT outcomes and support clinical decision-making.