基于深度学习的宫颈癌近距离治疗插植针重建研究  

Deep learning-based automatic reconstruction of interstitial needles in brachytherapy for cervical cancer

作　　者：文诗靖 刘涛 王思琪 徐立鹏 张庆贤[1] 王先良[2] Wen Shijing;Liu Tao;Wang Siqi;Xu Lipeng;Zhang Qingxian;Wang Xianliang(Applied Nuclear Technology in Geosciences Key Laboratory of Sichuan Province,Chengdu University of Technology,Chengdu 610059,China;Department of Radiation Oncology,Radiation Oncology Key Laboratory of Sichuan Province,Sichuan Clinical Research Center for Cancer,Sichuan Cancer Hospital&Institute,Sichuan Cancer Center,Affiliated Cancer Hospital of University of Electronic Science and Technology of China,Chengdu,610041,China)

机构地区：[1]成都理工大学核技术与自动化工程学院,成都610059 [2]放射肿瘤学四川省重点实验室,四川省肿瘤临床医学研究中心,四川省肿瘤医院·研究所,四川省癌症防治中心,电子科技大学附属肿瘤医院放疗科,成都610041

出　　处：《中华放射肿瘤学杂志》2025年第3期282-288,共7页Chinese Journal of Radiation Oncology

基　　金：四川省肿瘤医院优秀青年基金(YB2024004);四川省卫生健康委员会科技项目(24QNMP037);中央高校基本科研业务费专项资金(ZYGX2021YGCX002);辐射物理及技术教育部重点实验室开放课题(2023SCURPT05)。

摘　　要：目的探究基于深度学习实现宫颈癌腔内联合组织间插植近距离治疗(IC-ISBT)中插植针自动分割重建的可行性。方法回顾性收集接受IC-ISBT的98名患者的180次治疗计划数据,按照16∶1∶1的比例划分为训练集、验证集和测试集。使用放射源的驻留位置创建插植针的掩膜并训练3D U-Net模型。用Dice相似系数(DSC)评估预测模型性能,采用绝对准确率和相对准确率评估方法的整体效果,使用距离偏差评估横断面的定位精度,采用Wilcoxon秩和检验比较自动分割重建与人工重建的耗时差异,评估重建效率。结果预测模型的DSC为0.93±0.02,绝对准确率和相对准确率分别为0.44±0.09和0.95±0.03。距离偏差为(0.58±0.54)mm。自动分割重建耗时(6.2±0.4)s,显著小于人工重建的耗时(P<0.001)。结论基于深度学习模型,利用放射源的驻留位置进行数据标注,结合后处理算法可以对IC-ISBT的CT影像中的插植针进行精准的自动化分割重建,并显著提升重建效率。ObjectiveTo explore the feasibility of autosegmentation and reconstruction of interstitial needles in intracavitary/interstitial brachytherapy(IC-ISBT)for cervical cancer based on deep learning.MethodsThe data of 180 treatment plans from 98 patients who received IC-ISBT were retrospectively collected and divided into the training,validation,and testing sets in a 16:1:1 ratio.Masks of needles were created using the dwell positions of radiation sources,and a 3D U-Net model was trained.The performance of the model was evaluated using the Dice similarity coefficient(DSC).Absolute and relative accuracy rates were used to assess the results of this method,and the position bias was used to evaluate the precision of predictions in the transversal plan of CT scans.Wilcoxon rank-sum test was performed to evaluate the reconstruction efficiency by comparing the time required for automated versus manual reconstruction.ResultsDSC of the model was 0.93±0.02.The absolute and relative accuracy rates were 0.44±0.09 and 0.95±0.03,respectively.The distance deviation on the CT horizontal plane was(0.58±0.54)mm.The average time of autosegmentation and reconstruction was(6.2±0.4)s,leading to a significant reduction in time consumption compared with manual construction(P<0.001).ConclusionsBased on deep learning,using the dwell positions of radiation sources for data annotation,combined with post-processing algorithms,accurate automated segmentation and digital reconstruction of needles in IC-ISBT three-dimensional CT images can be achieved,significantly improving reconstruction efficiency.

关 键 词：深度学习 近距离放射疗法 宫颈肿瘤 插植针 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]