Value of CT radiomics for prediction of pathological response to neoadjuvant chemoradiotherapy in esophageal cancer
10.3760/cma.j.cn113030-20201116-00555
- VernacularTitle:基于多参数CT成像的食管癌新辅助放化疗病理反应预测
- Author:
Xiang ZHU
1
;
Chaonan ZHU
;
Jian ZENG
;
Xiaojiang SUN
;
Qingren LIN
;
Jun FANG
;
Ming CHEN
;
Yongling JI
Author Information
1. 中国科学院大学附属肿瘤医院（浙江省肿瘤医院）胸部放疗科，中国科学院基础医学与肿瘤研究所，浙江杭州 310022
- Keywords:
Radiomics;
Esophageal neoplasm/neoadjuvant chemoradiotherapy;
Pathological response
- From:
Chinese Journal of Radiation Oncology
2021;30(10):1019-1024
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To establish a radiomics-based biomarker for predicting pathological response after preoperative neoadjuvant chemoradiotherapy (nCRT) in locally advanced esophageal cancer.Methods:From 2008 to 2018, 112 patients with locally advanced esophageal cancer who received nCRT were enrolled. All patients were treated with preoperative nCRT combined with surgery. Enhanced CT images and clinical information before nCRT were collected. A lesion volume of interest was manually delineated. In total, 670 radiomics features (including tumor intensity, shape and size, texture and wavelet characteristics) were extracted using the pyradiomics package in PYTHON. The stepwise regression combined with the best subset were employed to select the features, and finally the Logistic regression model was adopted to establish the prediction model. The performance of the classifier was evaluated by the area under the ROC curve (AUC). Results:The pathological complete remission (pCR) rate was 58.0%(65/112). 10 radiomics features were included in the final model, The most relevant radiomics feature was the gray feature (the texture information of the image), followed by the shape and voxel intensity-related features. In the training set, the AUC was 0.750 with a sensitivity of 0.711 and a specificity of 0.778, the corresponding values in the testing set were 0.870, 0.757 and 0.900, respectively.Conclusions:Models based on radiomics features from CT images can be utilized to predict the pathological response to nCRT in esophageal cancer. As it is efficient, non-invasive and economic model, it could serve as a promising tool for individualized treatment when validated by further prospective trials in the future.
