| 7 | 0 | 50 |
| 下载次数 | 被引频次 | 阅读次数 |
目的：探讨基于深度学习的自动分割方法在椎体转移瘤放疗靶区勾画中的可行性和准确性。方法：选取2023年6月至2025年6月孝感市中心医院87例椎体转移瘤患者的CT影像及结构文件，完成匿名化处理后构建数据集。采用完全随机化方法，按照7.7∶1将87例病例分为训练集77例和测试集10例。采用UNet-2D、UNet-3D、RESUNet-2D与RESUNet-3D 4种模型进行训练与测试。利用戴斯系数(DSC)、交并比(IoU)和豪斯多夫距离(HD)对分割性能进行评价，并进行UNet-2D、UNet-3D、RESUNet-2D、RESUNet-3D模型组间差异性比较。结果：UNet-2D、UNet-3D、RESUNet-2D、RESUNet-3D模型的DSC值分别为0.769±0.051、0.826±0.060、0.799±0.051和0.839±0.059，其中RESUNet-3D模型DSC值最高，UNet-2D与UNet-3D、RESUNet-3D比较差异均有统计学意义(t=-2.298、-2.892,P<0.05)。各模型的IoU值分别为0.627±0.067、0.707±0.088、0.668±0.072和0.727±0.084，其中RESUNet-3D模型表现最佳，UNet-2D与UNet-3D、RESUNet-3D比较差异均有统计学意义(t=-2.316、-2.960,P<0.05)。各模型的HD值分别为7.721±9.319、2.269±2.335、3.629±3.337和2.636±2.37，其中UNet-3D模型数值相对较优。结论：与2D网络相比，3D网络在椎体转移瘤放疗靶区自动勾画中表现出更高的分割精度和稳定性，其中RESUNet-3D网络的综合性能最佳。基于3D残差网络的自动勾画方法有望在临床放疗中辅助医生进行器官勾画，提升工作效率和标准化程度。
Abstract：Objective:To investigate feasibility and accuracy of auto-segmentation method based on deep learning in delineation for target region of radiotherapy for metastatic tumor of vertebral body.Methods:The computed tomography(CT) images and structural files of 87 patients with metastatic tumor of vertebral body at the Central Hospital of Xiaogan during June 2023 and June 2025 were selected.These data were used to construct data set after the anonymization processing was completed.The 87 patients were divided into training set(77 cases) and test set(10 cases) as the ratio of 7.7 to 1 by using a complete randomization method.Four models included UNet-2D,UNet-3D,RESUNet-2D,and RESUNet-3D were trained and tested.Segmentation performance was evaluated by using Dice Similarity Coefficient(DSC),Intersection over Union(IoU),and Hausdorff Distance(HD),and the difference among them was compared.Results:The Dice values of UNet-2D,UNet-3D,RESUNet-2D,and RESUNet-3D were 0.769±0.051,0.826±0.060,0.799±0.051,and 0.839±0.059,respectively,and RESUNet-3D achieved the highest DSC value,and the differences of that between UNet-2D and UNet-3D,and between UNet-2D and RESUNet-3D were significant(t=-2.298,-2.892,P<0.05).The IoU values of all models were respectively 0.627±0.067,0.707±0.088,0.668±0.072,and 0.727±0.084,with RESUNet-3D model again performing best,and the differences of that between UNet-2D and UNet-3D,and between UNet-2D and RESUNet-3D were significant(t=-2.316,-2.960,P<0.05).The HD values of all models were respectively 7.721±9.319,2.269±2.335,3.629±3.337 and2.636±2.370,with UNet-3D showing relatively better performance.Conclusion:Compared with two-dimensional networks,threedimensional networks superior segmentation accuracy and stability in automatic delineation for target region of radiotherapy for metastatic tumor of vertebral body.Among of them,the RESUNet-3D model showed the best comprehensive performance.These findings suggest that 3D residual network-based automatic delineation method can assist clinicians to conduct delineation for organs in clinical radiotherapy,which can improve efficiency and standardization of radiotherapy.
[1]Wagner A,Haag E,Joerger AK,et al.Comprehensive surgical treatment strategy for spinal metastases[J].Sci Rep,2021,11(1):7988.DOI:10.1038/s41598-021-87121-1.
[2]Ma CY,Zhou JY,Xu XT,et al.Deep learning-based autosegmentation of clinical target volumes for radiotherapy treatment of cervical cancer[J].J Appl Clin Med Phys,2022,23(2):e13470.DOI:10.1002/acm2.13470.
[3]Liu ZK,Liu X,Guan H,et al.Development and validation of a deep learning algorithm for auto-delineation of clinical target volume and organs at risk in cervical cancer radiotherapy[J].Radiother Oncol,2020,153:172-179.DOI:10.1016/j.radonc.2020.09.060.
[4]Alomar K,Aysel HI,Cai X.Data augmentation in classification and segmentation:a survey and new strategies[J].J Imaging,2023,9(2):46.DOI:10.3390/jimaging9020046.
[5]Goceri E.Medical image data augmentation:techni ques,comparisons and interpretations[J].Artif Intell Rev,2023,56(11):12561-12605.DOI:10.1007/s10462-023-10453-z.
[6]Ronneberger O,Fischer P,Brox T.U-Net:Convolutional Networks for Biomedical Image Segmentation[C]//Medical Image Computing and Computer-Assisted Intervention(MICCAI).Cham:Springer,2015:234–241.DOI:10.1007/978-3-319-24574-4_28.
[7]Zhang Z,Liu Q,Wang Y.Road Extraction by Deep Residual U-Net[J]. IEEE Geosci Remote Sens Lett,2018,15(5):749–753.DOI:10.1109/LGRS.2018.2802944.
[8]He K,Zhang X,Ren S,et al.Deep Residual Learning for Image Recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR)Las Vegas,NV,USA:2016:770–778.DOI:10.1109/CVPR.2016.90.
[9]Li Z , Yang L , Shu L , et al. Research on CT lung segmentation method of preschool children based on traditional image processing and ResUnet[J].Comput Math Methods Med,2022,2022(1):7321330.DOI:10.1155/2022/7321330.
[10]He K,Zhang X,Ren S,et al.Identity mappings in deep residual networks[C]//European Conference on Computer Vision(ECCV).Cham:Springer,2016:630-645.DOI:10.1007/978-3-319-46493-038.
[11]Milletari F, Navab N , Ahmadi SA. V-net:Fully convolutional neural networks for volumetric medical image segmentation[C]//2016 Fourth International Conference on 3D Vision(3DV),IEEE,2016:565-571.DOI:10.1109/3DV.2016.79.
[12]Çiçe kÖ, Abdulkadir A , Lienkamp SS , et al. 3 DU-Net:learning dense volumetric segmentation from sparse annotation[C]//Medical Image Computing and Computer-Assisted Intervention(MICCAI).Cham:Springer,2016:424-432.DOI:10.1007/978-3-319-46723-8_49.
[13]陈英,张伟,林洪平,等.医学图像分割算法的损失函数综述[J].生物医学工程学杂志,2023,40(2):392-400.DOI:10.7507/1001-5515.202206038.
[14]Chen LC,Papandreou G,Kokkinos I,et al.Deeplab:semantic image segmentation with deep convolutional nets,atrous convolution,and fully connected crfs[J].IEEE Trans Pattern Anal Mach Intell,2017,40(4):834-848.DOI:10.1109/TPAMI.2017.2699184.
[15]陈科圻,朱志亮,邓小明,等.多尺度目标检测的深度学习研究综述[J].软件学报,2021,32(4):1201-1227.DOI:10.13328/j.cnki.jos.006166.
[16]Zhang C,Hua Q,Chu Y,et al.Liver tumor segmentation using 2.5 D UV-Net with multi-scale convolution[J].Comput Biol Med,2021,133:104424.DOI:10.1016/j.compbiomed.2021.104424.
[17]陈荣耀,吴乾健,黎美妍,等.人工智能在宫颈癌放射治疗中的应用进展[J].中国医学装备,2025,22(9):143-149.DOI:10.3969/j.issn.1672-8270.2025.09.028.
[18]陈超爽,曹洋森,朱晓斐,等.基于深度学习的胰腺肿瘤靶区自动分割[J].中国医学物理学杂志,2025,(7):923-928.DOI:10.3969/j.issn.1005-202X.2025.07.012.
[19]Alom MZ,Yakopcic C,Hasan M,et al.Recurrent residual U-Net for medical image segmentation[J].J Med Imaging,2019,6(1):014006-014006.DOI:10.1117/1.JMI.6.1.014006.
[20]Yang S,Jeong JS,Song D,et al.Comparison of 2D,2.5 D,and 3D segmentation networks for mandibular canals in CBCT images:a study on public and external datasets[J].BMC Oral Health,2025,25(1):1126.DOI:10.1186/s12903-023-03607-6.
[21]Kim TJ,Kim YJ,Kim KG,et al.Comparative analysis of brain tumor image segmentation performance of 2D U-Net and 3D U-Nets with alternative normalization methods[J].J Multimed Inf Syst,2024,11(2):157-166.DOI:10.33851/JMIS.2024.11.2.157.
基本信息:
中图分类号:R730.55;R738.1
引用信息:
[1]艾念,周雪阳,薄宏宇,等.基于残差U-net神经网络实现椎体转移瘤放疗靶区的自动勾画研究[J].中国医学装备,2026,23(05):28-32.
基金信息:
孝感市自然科学计划（XGKJ2021010044）~~
2026-05-25
2026-05-25