Xiaomiao Zhang1, Qi Zhang1, Xiaoduo Yu2, Xinming Zhao3, Yan Chen1, Sicong Wang4, Jieying Zhang1, Jusheng An5, Lizhi Xie4. 1. Department of Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100021, China. 2. Department of Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100021, China. yxd136@139.com. 3. Department of Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100021, China. zhaoxinming@cicams.ac.cn. 4. GE Healthcare, MR Research, Beijing, China. 5. Department of Gynecologic Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100021, China.
Abstract
PURPOSE: To investigate the value of magnetic resonance imaging (MRI)-based radiomics in predicting the treatment response to concurrent chemoradiotherapy (CCRT) in patients with locally advanced cervical squamous cell cancer (LACSC). METHODS: In total, 198 patients (training: n = 138; testing: n = 60) with LACSC treated with CCRT between January 2014 and December 2019 were retrospectively enrolled in this study. Responses were evaluated by MRI and clinical data performed at one month after completion of CCRT according to RECIST standards, and patients were divided into the residual group and nonresidual group. Overall, 200 radiomics features were extracted from T2-weighted imaging and apparent diffusion coefficient maps. The radiomics score (Rad-score) was constructed with a feature selection strategy. Logistic regression analysis was used for multivariate analysis of radiomics features and clinical variables. The performance of all models was assessed using receiver operating characteristic curves. RESULTS: Among the clinical variables, tumor grade and FIGO stage were independent risk factors, and the areas under the curve (AUCs) of the clinical model were 0.741 and 0.749 in the training and testing groups. The Rad-score, consisting of 4 radiomics features selected from 200 radiomics features, showed good predictive performance with an AUC of 0.819 in the training group and 0.776 in the testing group, which were higher than the clinical model, but the difference was not statistically significant. The combined model constructed with tumor grade, FIGO stage, and Rad-score achieved the best performance, with an AUC of 0.857 in the training group and 0.842 in the testing group, which were significantly higher than the clinical model. CONCLUSION: MRI-based radiomics features could be used as a noninvasive biomarker to improve the ability to predict the treatment response to CCRT in patients with LACSC.
PURPOSE: To investigate the value of magnetic resonance imaging (MRI)-based radiomics in predicting the treatment response to concurrent chemoradiotherapy (CCRT) in patients with locally advanced cervical squamous cell cancer (LACSC). METHODS: In total, 198 patients (training: n = 138; testing: n = 60) with LACSC treated with CCRT between January 2014 and December 2019 were retrospectively enrolled in this study. Responses were evaluated by MRI and clinical data performed at one month after completion of CCRT according to RECIST standards, and patients were divided into the residual group and nonresidual group. Overall, 200 radiomics features were extracted from T2-weighted imaging and apparent diffusion coefficient maps. The radiomics score (Rad-score) was constructed with a feature selection strategy. Logistic regression analysis was used for multivariate analysis of radiomics features and clinical variables. The performance of all models was assessed using receiver operating characteristic curves. RESULTS: Among the clinical variables, tumor grade and FIGO stage were independent risk factors, and the areas under the curve (AUCs) of the clinical model were 0.741 and 0.749 in the training and testing groups. The Rad-score, consisting of 4 radiomics features selected from 200 radiomics features, showed good predictive performance with an AUC of 0.819 in the training group and 0.776 in the testing group, which were higher than the clinical model, but the difference was not statistically significant. The combined model constructed with tumor grade, FIGO stage, and Rad-score achieved the best performance, with an AUC of 0.857 in the training group and 0.842 in the testing group, which were significantly higher than the clinical model. CONCLUSION: MRI-based radiomics features could be used as a noninvasive biomarker to improve the ability to predict the treatment response to CCRT in patients with LACSC.
Authors: Elisabeth Chereau; Claire DE LA Hosseraye; Marcos Ballester; Laurie Monnier; Roman Rouzier; Emmanuel Touboul; Emile Daraï Journal: Anticancer Res Date: 2013-04 Impact factor: 2.480