基于机器学习预测儿童腹盆部CT器官剂量的可行性研究  

Machine Learning Applied to Organ Dose Prediction in Pediatric Abdomen-Pelvis CT

作　　者：张欣欣 杨楠[1] 冯伦伦 高望 王春祥[1] ZHANG Xinxin;YANG Nan;FENG Lunlun(Department of Radiology,Tianjin Children’s Hospital(Children’s Hospital,Tianjin University),Tianjin 300134,P.R.China)

机构地区：[1]天津市儿童医院(天津大学儿童医院)医学影像科,天津市儿童出生缺陷防治重点实验室,300134

出　　处：《临床放射学杂志》2025年第4期712-717,共6页Journal of Clinical Radiology

基　　金：天津市医学重点学科(专科)建设项目基金资助项目(编号:TJYXZDXK-040A)。

摘　　要：目的 探究机器学习方法用于预测儿童腹盆部CT器官剂量的可行性。方法 回顾性搜集于天津市儿童医院接受腹盆部CT平扫的3415例患儿图像,选择年龄、性别、CT容积剂量指数(CTDIvol)、扫描长度、剂量长度乘积(DLP)和体型特异性剂量估计(SSDE)等六种特征,分别使用支持向量机(SVM)、决策树(DT)、随机森林(RF)和K近邻(KNN)等算法预测个体器官剂量,比较四种模型的性能。利用权重系数和Shapley加法解释(SHAP)确定每个特征对模型的贡献程度。结果 RF整体性能最优,其平均绝对误差、平均绝对百分比误差、均方误差、均方根误差、Pearson相关系数(r)和决定系数(R^(2))分别为1.6341、0.0218、7.3453、2.7102、0.9671(P<0.001)和0.9330。SSDE对于模型预测贡献程度最高,与预测值为正相关;年龄贡献度仅次于SSDE,与预测值为负相关。结论 临床工作中可使用基于SSDE等指标构建的机器学习模型准确快捷地估测个体器官剂量。Objective To explore the feasibility of machine learning techniques for predicting organ dose in pediatric abdomen-pelvis CT.Methods Images of 3415 children who underwent abdomen-pelvis CT scans in Tianjin Children's Hospital from July 2022 to December 2023 were retrospectively collected.Six features,including age,gender,volume Computed Tomography dose index(CTDIvol),scan length,dose-length product(DLP),and size-specific dose estimate(SSDE),were selected.Support vector machine(SVM),decision tree(DT),random forest(RF),and k-nearest neighbor(KNN) algorithms were used to predict individualized organ dose,and the performance of the four models was compared.The weight coefficient and Shapley additive explanations(SHAP) were used to determine the contribution of each feature to the model.Results The RF model had the best overall prediction performance.Its mean absolute error,mean absolute percentage error,mean square error,root mean square error,Pearsoncorrelation coefficientrand determination coefficient R~2 were 1.6341,0.0218,7.3453,2.7102,0.9671(P<0.001) and 0.9330,respectively.The SSDE contributed the most to model prediction,and it was positively correlated with the predicted value;the contribution of age was second only to SSDE,and was negatively correlated with the predicted value.Conclusion In clinical practice,machine learning models based on indicators such as SSDE can be used to accurately and quickly estimate individual organ doses.

关 键 词：器官剂量 体型特异性剂量估计 机器学习 儿童腹盆部CT 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]