CT-based radiomic signatures for prediction of pathologic complete response in esophageal squamous cell carcinoma after neoadjuvant chemoradiotherapy

• Published in: Journal of radiation research Vol. 60; no. 4; pp. 538 - 545
• Main Authors: Yang, Zhining, He, Binghui, Zhuang, Xinyu, Gao, Xiaoying, Wang, Dandan, Li, Mei, Lin, Zhixiong, Luo, Ren
• Format: Journal Article
• Language: English
• Published: England Oxford University Press 01.07.2019
• Subjects: Adult; Aged; Chemoradiotherapy, Adjuvant; Chemotherapy; Esophageal cancer; Esophageal Neoplasms - diagnostic imaging; Esophageal Neoplasms - drug therapy; Esophageal Neoplasms - radiotherapy; Esophageal Squamous Cell Carcinoma - diagnostic imaging; Esophageal Squamous Cell Carcinoma - drug therapy; Esophageal Squamous Cell Carcinoma - radiotherapy; Female; Humans; Likelihood ratio; Logistic Models; Male; Middle Aged; Parameters; Prediction models; Radiation therapy; Radiomics; Regression Analysis; Regression models; Regular Paper; Retrospective Studies; ROC Curve; Squamous cell carcinoma; Tomography, X-Ray Computed; Treatment Outcome; complete pathologic response; esophageal squamous cell carcinoma; neoadjuvant chemoradiotherapy; radiomics; LASSO
• DOI: 10.1093/jrr/rrz027
• ISSN: 0449-3060; 1349-9157

The objective of this study was to build models to predict complete pathologic response (pCR) after neoadjuvant chemoradiotherapy (nCRT) in esophageal squamous cell carcinoma (ESCC) patients using radiomic features. A total of 55 consecutive patients pathologically diagnosed as having ESCC were included in this study. Patients were divided into a training cohort (44 patients) and a testing cohort (11 patients). The logistic regression analysis using likelihood ratio forward selection was performed to select the predictive clinical parameters for pCR, and the least absolute shrinkage and selection operator (LASSO) with logistic regression to select radiomic predictors in the training cohort. Model performance in the training and testing groups was evaluated using the area under the receiver operating characteristic curves (AUC). The multivariate logistic regression analysis identified no clinical predictors for pCR. Thus, only radiomic features selected by LASSO were used to build prediction models. Three logistic regression models for pCR prediction were developed in the training cohort, and they were able to predict pCR well in both the training (AUC, 0.84–0.86) and the testing cohorts (AUC, 0.71–0.79). There were no differences between these AUCs. We developed three predictive models for pCR after nCRT using radiomic parameters and they demonstrated good model performance.