摘要
目的通过提取剂量图像影像组学特征建立机器学习模型,对患者容积调强放射治疗(VMAT)计划的伽马通过率结果进行分类。方法收集2020-01-01-2022-02-01在徐州市肿瘤医院放疗科行VMAT的523例患者治疗计划。其中,男267例,女256例;年龄22~87岁,中位年龄59岁。从每个三维剂量图像中提取107个影像组学特征,使用3%/2mm/10%阈值标准,建立随机森林分类模型。模型使用平均不纯度减少来选择重要特征,并对性能进行评估。结果根据Sklearn编程计算选择不同阈值下模型的精度,在阈值0.01下模型精度最高,筛选出了28个重要特征,最终模型在测试集的准确度为0.82,ROC曲线显示模型的曲线下面积为0.76。结论基于影像组学特征建立的随机森林模型可以对伽马通过率进行准确分类,对临床放射治疗质量保证工作具有一定的指导意义。
Objective To built a machine learning model to classify and predict the gamma pass rate results of patients'Volumetric Modulated Arc Therapy(VMAT)by extracting features from radiomics data of dose images.Methods Treatment plans were collected from 523clinical VMAT cases in the Department of Radiotherapy of Xuzhou Cancer Hospital from January 1,2020to February 1,2022.There were total 523patients including 267males,256females,and the median age was 59years(22-87years).A random forest classification and prediction model was built based on 107radiomics features which were extracted from each 3D dose image with a 3%/2mm/10%threshold criterion.The model was used to select significant features with average impurity reduction and the performance was evaluated.Results According to the Sklearn programming calculation,the model accuracy under different threshold values was selected,and the model accuracy under the threshold value of 0.01was the highest.The model screened out 28significant features with an accuracy of 0.82on the test set.The ROC curve showed an AUC of 0.76for the model.Conclusion The random forest model based on radiomics features can provide accurate classification results of gamma pass rates and guide work of clinical radiotherapy QA.
作者
陈宏林
梁冰花
陈晶晶
苗慧
CHEN Hong-lin;LIANG Bing-hua;CHEN Jing-jing;MIAO Hui(Department of Radiotherapy,Xuzhou Cancer Hospital,Xuzhou221005,China;Department of Radiotherapy,Pizhou People's Hospital,Pizhou221300,China)
出处
《中华肿瘤防治杂志》
CAS
北大核心
2022年第23期1697-1701,1708,共6页
Chinese Journal of Cancer Prevention and Treatment
基金
江苏省妇幼健康科研项目(F201950)。
关键词
剂量验证
伽马通过率
影像组学
分类模型
机器学习
dose verification
gamma pass rate
radiomics
classification model
machine learning