针对U-Net网络感受野受限以及信息丢失导致的分割精度低的问题,提出了一种基于感受野扩增和注意力机制的U-Net脑肿瘤MR图像分割算法.首先,在U-Net网络中引入感受野模块(receptive field block,RFB)来增大网络的感受野,解决了网络由于感受野受限带来的分割精度低的问题.此外在网络中引入有效的通道注意模块(efficient channel attention,ECA)来增加网络对有用特征的响应,抑制网络中的冗余特征.使用BraTS(the brain tumor image segmentation challenge)提供的脑肿瘤MR图像数据对本文算法进行测试,用Dice相似性系数等指标进行评价,结果显示在完整肿瘤、核心肿瘤以及增强肿瘤的Dice值分别可达到0.86、0.86、0.79.与U-Net模型以及其他的网络相比得到了提高.实验结果表明,本文提出的算法能够有效提升脑肿瘤分割的精度,具有良好的分割性能.
Although the current full convolutional neural network has been able to accurately identify the location of brain tumors in the brain tumor segmentation task, there are still problems of low segmentation accuracy due to limited receptive field and information loss. To solve these problems, this paper proposes a U-Net brain tumor MR image segmentation algorithm based on receptive field amplification and attentional mechanism. Firstly, the receptive field block(RFB)is introduced into the U-Net network to increase the receptive field of the network, which solves the problem of low segmentation accuracy caused by the limited receptive field. In addition, an efficient channel attention module(ECA)is introduced to increase the networks response to useful features and suppress redundant features in the network. The MR brain tumor image data provided by the BRATS(the brain tumor image segmentation- DOI:10.3969/j.issn.1000-1565.2022.05.013基于感受野扩增和注意力机制的U-Net脑肿瘤MR图像分割郑伟1,赵金芳1,张奕婧1,刘帅奇1,张晓丹2,马泽鹏2(1. 河北大学 电子信息工程学院,河北 保定 071002;2. 河北大学附属医院 超声诊断科,河北 保定 071000)摘 要:针对U-Net网络感受野受限以及信息丢失导致的分割精度低的问题,提出了一种基于感受野扩增和注意力机制的U-Net脑肿瘤MR图像分割算法.首先,在U-Net网络中引入感受野模块(receptive field block,RFB)来增大网络的感受野,解决了网络由于感受野受限带来的分割精度低的问题.此外在网络中引入有效的通道注意模块(efficient channel attention,ECA)来增加网络对有用特征的响应,抑制网络中的冗余特征.使用BraTS(the brain tumor image segmentation challenge)提供的脑肿瘤MR图像数据对本文算法进行测试,用Dice相似性系数等指标进行评价,结果显示在完整肿瘤、核心肿瘤以及增强肿瘤的Dice值分别可达到0.86、0.86、0.79.与U-Net模型以及其他的网络相比得到了提高.实验结果表明,本文提出的算法能够有效提升脑肿瘤分割的精度,具有良好的分割性能.关键词:脑肿瘤分割;U-Net;RFB;注意力机制 中图分类号:TN911 文献标志码:A 文章编号:1000-1565(2022)05-0542-10Segmentation algorithm of brain tumor MR images based on receptor field amplification and attentional mechanismZHENG Wei1, ZHAO Jinfang1, ZHANG Yijing1, LIU Shuaiqi1, ZHANG Xiaodan2, MA Zepeng2(1. College of Electronic and Information Engineering, Hebei University, Baoding 071002, China; 2. Department of Ultrasonography, Affilicated Hospital of Hebei University, Baoding 071000, China )Abstract: Although the current full convolutional neural network has been able to accurately identify the location of brain tumors in the brain tumor segmentation task, there are still problems of low segmentation accuracy due to limited receptive field and information loss. To solve these problems, this paper proposes a U-Net brain tumor MR image segmentation algorithm based on receptive field amplification and attentional mechanism. Firstly, the receptive field block(RFB)is introduced into the U-Net network to increase the receptive field of the network, which solves the problem of low segmentation accuracy caused by the limited receptive field. In addition, an efficient channel attention module(ECA)is introduced to increase the networks response to useful features and suppress redundant features in the network. The MR brain tumor image data provided by the BRATS(the brain tumor image segmentation- 收稿日期:2021-06-11 基金项目:河北省自然科学基金资助项目(F2020201025;H2020201021);河北省高等学校科学技术研究项目(BJ2020030) 第一作者:郑伟(1972—),女,黑龙江兰西人,河北大学教授,博士,主要从事图像处理与分析、图像安全通信、图像加密和隐藏方向研究.E-mail:147685650@qq.com 通信作者:张晓丹(1989—),女,河北承德人,河北大学附属医院主治医师,主要从事常规疾病及肿瘤病变的影像诊断.E-mail:zxd201505@126.com第5期郑伟等:基于感受野扩增和注意力机制的U-Net脑肿瘤MR图像分割challenge)was used to test the method, and indexes such as Dice similarity coefficient were used to evaluate the method. The results showed that the Dice of intact tumor, core tumor and enhanced tumor could reach 0.86, 0.86 and 0.79 respectively. Compared with the U-Net model and other networks, this method is improved. The results show that the proposed method can effectively improve the accuracy of brain tumor segmentation and has good segmentation performance.
[1] FITZMAURICE C, ALLEN C, BARBER R M, et al. Global, regional, and national cancer incidence, mortality, years of life lost, years lived with disability, and disability-adjusted life-years for 32 cancer groups, 1990 to 2015: a systematic analysis for the global burden of disease study[J]. JAMA Oncology, 2017, 3(4): 524-548. DOI:10.1001/jamaoncol.2016.5688.
[2] 禇晶辉,李晓川,张佳琪,等.一种基于级联卷积网络的三维脑肿瘤精细分割[J]. 激光与光电子学进展, 2019, 56(10): 67-76. DOI:10.3788/LOP56.101001.
[3] LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C] //Proceedings of the IEEE conference on computer vision and pattern recognition, Boston, MA, USA: IEEE, 2015: 3431-3440. DOI:10.1109/CVPR.2015.7298965.
[4] RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation[C] // International Conference on Medical image computing and computer-assisted intervention, Berlin, Germany: Springer Shop, 2015: 234-241. DOI: 10.1007/978-3-319-24574-4_28.
[5] HAO D, YANG G, LIU F, et al. Automatic brain tumor detection and segmentation using U-Net based fully convolutional networks[C] //annual conference on medical image understanding and analysis, Berlin, Germany: Springer Shop, 2017: 506-517. DOI:10.1007/978-3-319-60964-5_44.
[6] DI L, HAO Z, ZHAO M, et al. Brain tumor segmention based on dilated convolution refine networks[C] // IEEE 16th International Conference on Software Engineering Research, Management and Applications(SERA), Boston, MA, USA: IEEE, 2018: 113-120. DOI:10.1109/SERA.2018.8477213.
[7] LIU S, HUANG D, WANG Y. Receptive field block net for accurate and fast object detection[C] //Proceedings of the European conference on computer vision(ECCV), Berlin, Germany: Springer Shop, 2018: 385-400. DOI:10.1007/978-3-030-01252-6_24.
[8] WANG Q, WU B, ZHU P, et al. ECA-Net: Efficient channel attention for deep convolutional neural networks[C] //IEEECVF Conference on Computer Vision and Pattern Recognition(CVPR),Boston, MA, USA: IEEE, 2020:11531-11539. DOI:10.1109/CVPR42600.2020.01155.
[9] SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C] //Proceedings of the IEEE conference on computer vision and pattern recognition(CVPR), Boston, MA, USA: IEEE, 2016: 2818-2826. DOI:10.1109/CVPR.2016.308.
[10] CHEN L C, PAPANDREOU G, KOKKINOS I, et al. DeepLab: Semantic image iegmentation with deep convolutional nets, atrous convolution, and fully connected CRFs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(4):834-848. DOI:10.1109/TPAMI.2017.2699184.
[11] CHEN L C, PAPANDREOU G, KOKKINOS I, et al. Deeplab: sSemantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 40(4): 834-848. DOI:10.1109/TPAMI.2017.2699184.
[12] WOO S, PARK J, LEE J Y, et al. Cbam: Convolutional block attention module[C] //Proceedings of the European Conference on Computer Vision(ECCV), Berlin, Germany: Springer Shop, 2018: 3-19. DOI:10.1007/978-3-030-01234-2_1.
[13] JIE H, LI S, GANG S, et al. Squeeze-and-Excitation Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011-2023. DOI:10.1109/TPAMI.2019.2913372.
[14] BJOERN H M, REYES M, LEEPUT K V, et al. The multimodal brain tumor image segmentation benchmark(BRATS)[J]. IEEE Transactions on Medical Imaging, 2015, 34(10):1993-2024. DOI:10.1109/TMI.2014.2377694. (
