中国医学装备2024，Vol.21Issue(4)：7-12,27,7.DOI:10.3969/j.issn.1672-8270.2024.04.002
基于深度学习的新型妇科后装施源器自动重建系统研发
Research and development of a novel automatic reconstruction system based on deep learning for gynecological applicator
摘要
Abstract
Objective:We have developed a deep learning-based automatic reconstruction system for applicators,aiming to achieve efficient and accurate automatic reconstruction of Fletcher applicators in computed tomography(CT)-guided gynecological brachytherapy.Methods:The automatic reconstruction system of applicator was divided into two main parts.One part was an applicator mask on CT images that was split by 2.5D DpnUNet,and other part was a digitized centerline of the applicator channel that was obtained through three-dimensional(3D)connected region algorithm and skeleton extraction algorithm.The documents of 68 patients who received gynecological brachytherapy in Peking Union Medical College Hospital from July of 2022 to July of 2023 were selected.The CT plans of 10 patients of them were used as test set,and the CT plans of other 58 patients were used to train by adopting 10-fold cross validation method.The comparison of geometric consistency was conducted between the results of developed automatic reconstruction system and the results of manual reconstruction.The high risk clinical target volume(HR-CTV),90%and 98%dose target volume(D90,D98)of dosimetric indicators,as well as the minimum dose(D2cc)within 2cc volume that received maximum exposure dose on bladder,rectum,sigmoid colon and small intestine,were designed and obtained through the 3D rear mounted reverse optimization plan.And then,the clinical feasibility of the automatic reconstruction system was further explored.Results:In the data of 10 patients of test set,the average distances of automatic reconstruction,manual reconstruction and the top-end of the centerline of left-right fornix canal were respectively 0.335,0.361 and 0.362 mm.The average Hausdorff distances between the centerlines were respectively 0.398,0.367 and 0.324 mm.Additionally,the differences of dose-volume histogram(DVH)parameters between the two types of plans was less than 2%under kept the consistency between location and duration of stay.There were very high geometric consistency and clinical value between the two types of plans.Conclusion:The automatic reconstruction system of applicators can realize fully-automatic reconstruction with high-precision of Fletcher applicators,and reduce the probability of potential human error and improve clinical work efficiency.关键词
深度学习/近距离治疗/妇科施源器Key words
Deep learning/Brachytherapy/Gynecological applicator分类
医药卫生引用本文复制引用
张文君,于浪,张杰,杨波,罗春丽,邱杰..基于深度学习的新型妇科后装施源器自动重建系统研发[J].中国医学装备,2024,21(4):7-12,27,7.基金项目
中央高水平医院临床科研项目近距离放射治疗创新技术的研发和临床应用(2022-PUMCH-B-052) （2022-PUMCH-B-052）
北京协和医院中央高水平医院临床科研专项2022年青年培优计划项目(2022-PUMCH-A-101) Central High-level Hospital Clinical Research Project on the Research and Clinical Application of Innovative Technologies in Brachytherapy(2022-PUMCH-B-052) （2022-PUMCH-A-101）
Peking Union Medical College and the National High Level Hospital Clinical Research Funding(2022-PUMCH-A-101) （2022-PUMCH-A-101）
