Applying radiomics and dosimetry features to predict 2-year survival of esophageal cancer patients treated with radiotherapy
10.3969/j.issn.1000-8179.2020.07.312
- VernacularTitle: 应用放射组学和剂量学特征预测食管癌放射治疗后的两年生存情况*
- Author:
Chengwen YANG
1
Author Information
1. Department of Radiation Oncology, Tianjin Medical University Cancer Institute and Hospital, National Clinical Research Center for Cancer, Tianjin Key Laboratory of Cancer Prevention and Therapy, Tianjin's Clinical Research Center for Cancer
- Publication Type:Journal Article
- Keywords:
2-year survival;
Dosimetry characteristics;
Esophageal cancer;
Multi-parameter predictive model;
Radiomics;
Radiotherapy
- From:
Chinese Journal of Clinical Oncology
2020;47(7):334-337
- CountryChina
- Language:Chinese
-
Abstract:
Objective: Applying radiomics and dosimetry features to establish machine learning models, which is used to predict the 2-year survival of esophageal patients with radiotherapy. Methods: Retrospective analysis of 579 esophageal cancer patients who underwent radiotherapy from January 2013 to December 2017 in Tianjin Medical University Cancer Institute and Hospital. Radiomics and dosimetry features were extracted from the GTV of the radiotherapy plan for patients with esophageal cancer. The maximum correlation and minimum redundancy and manual methods were used to reduce the feature vector. A total of 14 radiomics and 14 dosimetry features were selected, then normalized to the range [0,1]. The machine learning models such as support vector machines (SVM), Logistic regression (LR), and random forest (RF) were used to train and test the radiomics and dosimetry features, respectively, then to predict the 2-year survival of esophageal cancer patients treated with radiotherapy. Results: When only the radiomics features were used to predict the 2-year survival after radiotherapy, the accuracy of SVM, LR and RF models were 84.98%, 85.92% and 84.51%, respectively. Furthermore, when the combined features of radiomics and dosimetry were used for prediction, the accuracy of the SVM, LR and RF models were 86.32%, 83.02% and 90.01%, respectively. Using the radiomics and dosimetry features, the predictive accuracy of SVM and RF models are effectively improved. Conclusions: For the SVM and RF models, the radiomics and dosimetry features can effectively improve the accuracy of predicting 2-year survival for esophageal cancer patients after radiation therapy.