基于深度学习的胸部CT图像自动化水当量直径(WED)估算与辐射剂量评估研究  

Research on Automated Water Equivalent Diameter(WED)Estimation and Radiation Dose Assessment of Chest CT Images Based on Deep Learning

作　　者：李欣颖 李玲[1] 张晓东[1] 刘建新[1] LI Xinying;LI Ling;ZHANG Xiaodong;LIU Jianxin(Department of Radiology,Peking University First Hospital,Beijing 100034,China)

机构地区：[1]北京大学第一医院医学影像科,北京100034

出　　处：《影像科学与光化学》2025年第2期134-139,共6页Imaging Science and Photochemistry

摘　　要：目的:开发并验证一种基于深度学习模型的全自动方法,用于从常规胸部CT图像中精确估算WED,以提高SSDE的评估效率和准确性。方法:回顾性收集了本院连续1500例患者的胸部CT图像数据,利用3D深度残差网络(deepresidual network,ResNet)与UNet网络结合的深度学习模型自动分割胸部扫描区域,并计算全扫描范围内的WED。通过DICE系数评估分割性能,并以中间层估算法和有效直径估算法比较WED估算的准确性。结果:在测试集中,模型分割性能的DICE系数为0.996。与中间层方法(误差百分比4.53%)和有效直径方法(误差百分比12.84%)相比,全扫描范围方法深度学习模型估算的WED(25.11±2.31 cm)具有更高的准确性。结论:本研究提出的深度学习模型实现了胸部CT图像WED的全自动精确估算,为多种临床辐射剂量评估提供了可靠工具,有助于优化CT检查方案,确保患者安全。Objective:To develop and validate a fully automated method based on deep learning models to accurately estimate WED from conventional chest CT images,in order to improve the efficiency and accuracy of SSDE evaluation.Methods:Retrospective collection of chest CT image data from 1500 consecutive patients in our hospital was conducted.A deep learning model combining 3D deepresidual network(ResNet)and UNet network was used to automatically segment the chest scan area and calculate the WED within the full scan range.Evaluate segmentation performance through DICE coefficient and compare the accuracy of WED estimation with intermediate layer estimation method and effective diameter estimation method.Results:In the test set,the DICE coefficient of the model segmentation performance was 0.996.Compared with the intermediate layer method(error percentage 4.53%)and the effective diameter method(error percentage 12.84%),the deep learning model of the full scan range method had higher accuracy in estimating WED(25.11±2.31)cm.Conclusion:The deep learning model proposed in this study achieves fully automatic and accurate estimation of WED in chest CT images,providing a reliable tool for various clinical radiation dose assessments,helping to optimize CT examination plans and ensure patient safety.

关 键 词：体型特异性剂量估计 水当量直径 分割 深度学习 

分 类 号：R‐1[医药卫生]