Adaptable Global Network for Whole-Brain Segmentation with Symmetry Consistency Loss | |
Zhao, Yuan-Xing3,4; Zhang, Yan-Ming3; Song, Ming2,3; Liu, Cheng-Lin1,3,4 | |
刊名 | COGNITIVE COMPUTATION |
2022-05-31 | |
页码 | 14 |
关键词 | Whole-brain segmentation Adaptable global network Semi-supervised learning Symmetry consistency loss |
ISSN号 | 1866-9956 |
DOI | 10.1007/s12559-022-10011-9 |
通讯作者 | Zhang, Yan-Ming(ymzhang@nlpr.ia.ac.cn) |
英文摘要 | Segmenting the whole brain into a large number (for example, >= 100) of regions is challenging due to the complexity of the brain and the lack of annotated data. Deep neural network-based segmentation methods have shown promise, but due to the limitation of graphics processing unit (GPU) memory, they cannot fully exploit the brain structure information contained in 3D data. This paper proposes a memory-efficient framework to exploit the global brain structure for whole-brain segmentation. In this framework, upon extracting the brain region by using a skull-stripping subnetwork, a global modeling subnetwork is used to learn a global brain representation for segmentation, while an adaptable segmentation subnetwork is used to optimize the global representation during training and directly segment the whole brain during testing. This framework enables the representation to be learned from the global structure with reduced memory consumption, and segmentation is performed without splitting the brain into patches. To overcome the lack of annotated data, we also propose a semi-supervised method based on a symmetry consistency loss and a prior knowledge- based pseudolabel generation strategy. Extensive experiments on four datasets demonstrate that our method outperforms previously developed methods and achieves state-of-the-art performance. The method is computationally efficient in that segmenting a raw magnetic resonance imaging (MRI) image requires less than 2 s on a TITAN X GPU; our approach is much faster than multiatlas-based methods and previously proposed 3D deep learning methods. The code is publicly available at https://github.com/ZYX-MLer/AGNetwork. |
资助项目 | National Key Research and Development Program[2018AAA0100400] ; National Natural Science Foundation of China (NSFC)[61773376] ; National Natural Science Foundation of China (NSFC)[61836014] ; National Natural Science Foundation of China (NSFC)[61721004] ; National Natural Science Foundation of China (NSFC)[31870984] |
WOS关键词 | CONVOLUTIONAL NEURAL-NETWORKS |
WOS研究方向 | Computer Science ; Neurosciences & Neurology |
语种 | 英语 |
出版者 | SPRINGER |
WOS记录号 | WOS:000803780800002 |
资助机构 | National Key Research and Development Program ; National Natural Science Foundation of China (NSFC) |
内容类型 | 期刊论文 |
源URL | [http://ir.ia.ac.cn/handle/173211/49554] |
专题 | 自动化研究所_模式识别国家重点实验室_模式分析与学习团队 |
通讯作者 | Zhang, Yan-Ming |
作者单位 | 1.Ctr Excellence Brain Sci & Intelligence Technol, Beijing 100049, Peoples R China 2.Chinese Acad Sci, Brainnetome Ctr, Inst Automat, Beijing 100049, Peoples R China 3.Chinese Acad Sci, Natl Lab Pattern Recognit, Inst Automat, Beijing 100049, Peoples R China 4.Univ Chinese Acad Sci, Sch Artificial Intelligence, Beijing 100049, Peoples R China |
推荐引用方式 GB/T 7714 | Zhao, Yuan-Xing,Zhang, Yan-Ming,Song, Ming,et al. Adaptable Global Network for Whole-Brain Segmentation with Symmetry Consistency Loss[J]. COGNITIVE COMPUTATION,2022:14. |
APA | Zhao, Yuan-Xing,Zhang, Yan-Ming,Song, Ming,&Liu, Cheng-Lin.(2022).Adaptable Global Network for Whole-Brain Segmentation with Symmetry Consistency Loss.COGNITIVE COMPUTATION,14. |
MLA | Zhao, Yuan-Xing,et al."Adaptable Global Network for Whole-Brain Segmentation with Symmetry Consistency Loss".COGNITIVE COMPUTATION (2022):14. |
个性服务 |
查看访问统计 |
相关权益政策 |
暂无数据 |
收藏/分享 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论