AUTHOR=Xi Yuzhen , Ge Xiuhong , Ji Haiming , Wang Luoyu , Duan Shaofeng , Chen Haonan , Wang Mengze , Hu Hongjie , Jiang Feng , Ding Zhongxiang 

TITLE=Prediction of Response to Induction Chemotherapy Plus Concurrent Chemoradiotherapy for Nasopharyngeal Carcinoma Based on MRI Radiomics and Delta Radiomics: A Two-Center Retrospective Study

JOURNAL=Frontiers in Oncology

VOLUME=Volume 12 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2022.824509

DOI=10.3389/fonc.2022.824509

ISSN=2234-943X

ABSTRACT=Objective: We aimed to establish a radiomics model to predict tumor retraction after induction chemotherapy combined with concurrent chemoradiation (CCRT) for primary nasopharyngeal carcinoma (NPC) in non-endemic areas based on MRI images before treatment and after induction chemotherapy (IC) and to validate its efficacy.
Methods: A total of 668 patients with pathologically confirmed primary NPC who underwent biopsy at the Cancer Hospital of the University of Chinese Academy of Sciences (Zhejiang Cancer Hospital) between January 2007 and June 2012 were retrospectively included. All 221 patients (155 in the training set and 66 in the internal validation set) before treatment and 96 patients (67 in the training set and 28 in the internal validation set) after IC were screened for clinical response after IC combined with CCRT (IC-CCRT) according to the Response Evaluation Criteria in Solid Tumors 1.1 (RECIST). Fifty-one patients with primary NPC at the Sir Run Run Shaw Hospital, Zhejiang University were also included for external validation. After maximum relevance minimum redundancy (mRMR) and LASSO algorithms, logistic regression models were established, optimal Youden index was taken, ROC curve, calibration curve, and decision curve were drawn to evaluate the predictive efficacy of the two different models. 
Results: Seven optimal feature subsets were selected from the pretreatment MRI model, and six optimal subsets were selected from the IC MRI model. The area under the ROC curve, accuracy, sensitivity, specificity, NPV, and PPV of the former were 0.865, 0.827, 0.837, 0.813, 0.776, and 0.865, respectively; the corresponding indicators of the latter were 0.865, 0.821, 0.694, 0.968, 0.732, and 0.962, respectively.
Conclusion: The MRI radiomics model can predict the IC-CCRT response of NPC in non-epidemic areas.