Selection of respiratory motion management strategies for stereotactic radiotherapy in liver cancer based on machine learning
10.3760/cma.j.cn113030-20240703-00258
- VernacularTitle:基于机器学习的肝癌立体定向放疗呼吸运动管理方式选择
- Author:
Shiqin DENG
1
;
Zhen YANG
;
Du TANG
;
Hua PENG
;
Zhao PENG
;
Ying CAO
;
Xiaoyu YANG
;
Shuzhou LI
;
Kan CHEN
Author Information
1. 中南大学湘雅医院肿瘤科，长沙 410008
- Publication Type:Journal Article
- Keywords:
Liver neoplasms;
Stereotactic body radiotherapy;
Respiratory motion evaluation;
Abdominal compression;
Free breathing
- From:
Chinese Journal of Radiation Oncology
2025;34(4):363-368
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the feasibility of selecting appropriate respiratory motion management strategies for patients undergoing stereotactic radiotherapy for liver cancer using chi-square feature testing and machine learning techniques.Methods:A retrospective analysis was conducted on 95 liver cancer patients who underwent respiratory motion evaluation at Xiangya Hospital of Central South University between March 2022 and August 2024. Chi-square testing was used to screen features related to respiratory motion evaluation in liver cancer patients. Based on these features, predictive models were constructed using 4 machine learning classification algorithms: support vector machine (SVM), random forest (RF), gradient boosting decision tree (GBDT), and adaptive boosting (AdaBoost). The predictive performance of these models was evaluated using accuracy, sensitivity, specificity, F1 value, and area under the curve (AUC).Results:The accuracy values for the SVM, RF, GBDT and AdaBoost models were 0.75, 0.75, 0.70, and 0.82, respectively. The sensitivity values were 0.82, 0.82, 0.64, and 0.82, respectively. The specificity values were 0.63, 0.63, 0.63, and 0.75, respectively. The F1 scores were 0.78, 0.78, 0.67, and 0.82, respectively. The AUC values were 0.85, 0.80, 0.76, and 0.85, respectively.Conclusions:The predictive models constructed by combining chi-square feature testing and machine learning methods can effectively predict the selection of respiratory motion management strategies. Among the models, the AdaBoost model demonstrated the best predictive performance for selecting respiratory motion management strategies.
