Self-supervised Edge-Fusion Network for MRI Reconstruction
LI Zhongnian1, ZHANG Tao1, ZHANG Daoqiang1
1. MIIT Key Laboratory of Pattern Analysis and Machine Intelligence, College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 211100
Abstract：The research on compressed sensing magnetic resonance imaging(CS-MRI) suggests that the edge information is the hardest part of medical image reconstruction. In most deep-learning based methods, the explicit consideration for edge information is not taken into account. To tackle this problem, a self-supervised edge-fusion network(SEN) is proposed to explore beneficial edge properties to reconstruct MRI. Firstly, edge annotations are generated by utilizing canny edge detector without involving any time-consuming and expensive human labeling. Secondly, a self-supervised auxiliary network is introduced to incorporate edge annotations into a feature learning to capture fusible representations. A top-down fusion strategy is proposed to fuse the learned representations into reconstruction network for CS-MRI restoring. Experimental results show that SEN catches the edge information effectively and achieves better performance in CS-MRI reconstruction.
[1] LUSTIG M, DONOHO D L, SANTOS J M, et al. Compressed Sen-sing MRI. IEEE Signal Processing Magazine, 2008, 25(2): 72-82. [2] CANDÈS E J, ROMBERG J, TAO T. Robust Uncertainty Principles: Exact Signal Reconstruction from Highly Incomplete Frequency Information. IEEE Transactions on Information Theory, 2006, 52(2): 489-509. [3] SUN L Y, FAN Z W, DING X H, et al. Joint CS-MRI Reconstruction and Segmentation with a Unified Deep Network // Proc of the International Conference on Information Processing in Medical Imaging. Berlin, Germany: Springer, 2019: 492-504. [4] QIN C, SCHLEMPER J, CABALLERO J, et al. Convolutional Recurrent Neural Networks for Dynamic MR Image Reconstruction. IEEE Transactions on Medical Imaging, 2019, 38(1): 280-290. [5] QU X B, HOU Y K, LAM F, et al. Magnetic Resonance Image Reconstruction from Undersampled Measurements Using a Patch-Based Nonlocal Operator. Medical Image Analysis, 2014, 18(6): 843-856. [6] RONNEBERGER O, FISCHER P, BROX T. U-Net: Convolutional Networks for Biomedical Image Segmentation // Proc of the International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin, Germany: Springer, 2015: 234-241. [7] YANG G, YU S M, DONG H, et al. DAGAN: Deep De-aliasing Generative Adversarial Networks for Fast Compressed Sensing MRI Reconstruction. IEEE Transactions on Medical Imaging, 2018, 37(6): 1310-1321. [8] SEITZER M, YANG G, SCHLEMPER J, et al. Adversarial and Perceptual Refinement for Compressed Sensing MRI Reconstruction // Proc of the International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin, Germany: Springer, 2018: 232-240. [9] NOROOZI M, FAVARO P. Unsupervised Learning of Visual Representations by Solving JIGSAW Puzzles // Proc of the European Conference on Computer Vision. Berlin, Germany: Springer, 2016: 69-84. [10] LIU X L, VAN DE WEIJER J, BAGDANOV A D. Leveraging Unlabeled Data for Crowd Counting by Learning to Rank // Proc of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2018: 7661-7669. [11] LARSSON G, MAIRE M, SHAKHNAROVICH G. Colorization as a Proxy Task for Visual Understanding // Proc of the IEEE Confe-rence on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2017: 6874-6883. [12] KINGMA D P, BA J M. Adam: A Method for Stochastic Optimization[C/OL]. [2020-05-25]. https://arxiv.org/pdf/1412.6980v1.pdf [13] HORÉ A, ZIOU D. Image Quality Metrics: PSNR vs. SSIM // Proc of the 20th International Conference on Pattern Recognition. Washington, USA: IEEE, 2010: 2366-2369.
2021, Vol. 34 
