| 基于循环一致性生成对抗网络的盆腔伪CT生成方法 |
| |
| 引用本文： | 吴香奕,曹锋,曹瑞芬,吴茜,董江宁,徐榭,裴曦. 基于循环一致性生成对抗网络的盆腔伪CT生成方法[J]. 中国医学物理学杂志, 2021, 0(1): 21-29. DOI: DOI:10.3969/j.issn.1005-202X.2021.01.005 |
| |
| 作者姓名： | 吴香奕 曹锋 曹瑞芬 吴茜 董江宁 徐榭 裴曦 |
| |
| 作者单位： | 1.中国科学技术大学物理学院工程与应用物理系， 安徽 合肥 230025； 2.中国科学技术大学第一附属医院影像科， 安徽 合肥 230001； 3.安徽大学计算机科学与技术学院， 安徽 合肥 230601； 4.安徽医科大学人文医学院， 安徽 合肥 230032； 5.安徽慧软科技有限公司， 安徽 合肥 230088 |
| |
| 基金项目： | 国家自然科学基金(11575180);安徽省自然科学基金(1908085MA27);安徽省重点研究与开发计划(1804a09020039);安徽省高校自然科学研究项目(KJ2019A0240)。 |
| |
| 摘 要： | 目的:基于循环一致性生成对抗网络（CycleGAN），利用非配对患者盆腔部位数据，实现MRI和CT图像之间的相互转换，并对基于该模型生成的盆腔伪CT（sCT）进行精度和剂量性能的评估。方法:该CycleGAN网络包含两个生成器和两个判别器。先基于全卷积网络（FCNs）构建两个生成器，一个将2D盆腔MRI转换为2D盆腔sCT图像，另一个将CT图像转换为伪MRI（sMRI）图像。再基于FCNs构建两个判别器，用于对真实图像和生成的伪图像进行判别，提升生成图像的质量。为保证sCT图像与MRI图像的一致性，引入归一化互信息作为相似性约束损失项，对模型进行改进。训练集包括35例患者盆腔部位的T1-MRI图像和另外36例患者盆腔部位的CT图像，测试集包括10例盆腔部位患者的MRI和CT图像，评估方法包括sCT与CT图像的误差和放疗剂量伽马通过率。结果:对于测试集中所有病例，生成的sCT与真实CT图像之间的平均绝对误差（MAE）为35.537（±4.537） HU；基于体素的平均剂量差异最大为0.49%；以3%/3 mm、2%/2 mm和1%/1 mm为标准的平均伽马通过率分别高于99%、98%和95%。结论:使用CycleGAN网络和非配对患者训练数据可以生成准确且符合临床剂量精度要求的盆腔部位sCT图像。
|
| 关 键 词： | 伪CT 非配对 盆腔 磁共振成像 循环一致性生成对抗网络 |
The method of pelvic synthetic CT generation based on the cycle-consistent generative adversarial networks |
| |
| WU Xiangyi,CAO Feng,CAO Ruifen,WU Qian,DONG Jiangning,X.George Xu,PEI Xi. The method of pelvic synthetic CT generation based on the cycle-consistent generative adversarial networks[J]. Chinese Journal of Medical Physics, 2021, 0(1): 21-29. DOI: DOI:10.3969/j.issn.1005-202X.2021.01.005 |
| |
| Authors: | WU Xiangyi CAO Feng CAO Ruifen WU Qian DONG Jiangning X.George Xu PEI Xi |
| |
| Affiliation: | 1. School of Physical Sciences, University of Science and Technology of China, Hefei, 230025, China 2. Department of Medical Imaging, the First Affiliated Hospital of University of Science and Technology of China, Hefei 230001, China 3. College of Computer Science and Technology, Anhui University, Hefei 230601, China 4. School of Humanistic Medicine, Anhui Medical University, Hefei 230032, China 5. Anhui Wisdom Technology Co. Ltd, Hefei 230088, China |
| |
| Abstract: | Objective Based on the Cycle-Consistent Generative Adversarial Networks(CycleGAN),pelvic data from unpaired patients are applied to achieve interconversion between MRI images and CT images,and the accuracy and dose performance of synthetic CT(s CT)images generated based on this model are assessed.Methods This Cycle GAN network includes two generators and two discriminators.Based on fully convolutional networks(FCNs),two generators were constructed first.One converted 2D pelvic MRI images to 2D pelvic s CT images,and the other converted CT images to synthetic MRI(s MRI)images.Based on FCNs,two discriminators were then constructed for discriminating real images and generated synthetic images to facilitate the s CT/s MRI quality improvement.To ensure the consistency between s CT images and MRI images,normalized mutual information was applied as a similarity-constraint loss term to improve the model.The training set consists of T1-weighted pelvic MRI images from 35 patients and pelvic CT images from an additional 36 patients,and the test set consists of pelvic MRI images and CT images from10 patients.Assessment methods included errors between sCT images and CT images and the radiation dose gamma pass rates.Results For all the cases in the test set,the mean absolute error(MAE)between the generated s CT images and the true CT images was 35.537(±4.537)HU.The maximum voxel-based mean dose difference was 0.49%.The mean gamma pass rates were above99%,98%and 95%with 3%/3 mm,2%/2 mm,and 1%/1 mm criteria,respectively.Conclusion Accurate pelvic s CT images can be generated by using CycleGAN network and unpaired training data from different patients,and the dose accuracy calculated based on the sCT images meets the clinical requirements. |
| |
| Keywords: | synthetic CT unpaired pelvis MRI CycleGAN |
| 本文献已被 CNKI 维普 等数据库收录！ |
| 点击此处可从《中国医学物理学杂志》浏览原始摘要信息 |
|
点击此处可从《中国医学物理学杂志》下载免费的PDF全文 |
|
