Fast EPRI Imaging Based on 3MNet Denoising Network
-
摘要: 电子顺磁共振成像(EPRI)是一种先进的氧成像技术。当前EPRI的瓶颈问题是扫描速度过慢,其原因是每个角度下的投影信号需要被重复采集几千次,以压制随机噪声。一种实现快速扫描的方法是减少投影信号的重复采集次数,然而这又使得投影信号信噪比降低,重建出来的图像噪声较大。为有效压制重建图像中的噪声,本文提出一种基于多通道、多尺度、多拼接的(3MNet)图像去噪网络,以实现高精度快速EPRI成像。该网络由3个子网络构成。第1个子网络是基于注意力机制的卷积网络,其输出的特征图像与输入图像拼接以构成后端网络的输入;第2个子网络是3通道卷积网络;第3个子网络是多尺度卷积网络。实验结果表明,本文提出的3MNet网络可以实现EPRI图像的高精度去噪,进而实现快速成像。Abstract: Electron paramagnetic resonance imaging (EPRI) is an advanced technique for oxygen imaging. The current bottleneck in EPRI is the slow scanning speed, due to the fact that the projection signal at each angle needs to be repeated thousands of times to suppress random noise. One way to achieve fast scanning is to reduce the number of repeated projection signal acquisitions; however, this, in turn, reduces the signal-to-noise ratio of the projection signal and results in a noisy reconstructed image. To effectively suppress the noise in the reconstructed image, this study proposes a multi-channel, multi-scale, multi-concatenation, and convolutional network (3MNet) based image denoising network to achieve high accuracy and fast EPRI imaging. The proposed network consists of three sub-networks. The first sub-network is an attention-based convolutional network, whose output feature images are stitched with the input images to form the input of the back-end network. The second sub-network is a three-channel convolutional network. Finally, the third sub-network is a multi-scale convolutional network. The experimental results demonstrate that the proposed 3MNet network can achieve high accuracy in denoising EPRI images and fast imaging.
-
Key words:
- deep learning /
- convolutional neural network /
- EPRI /
- rapid imaging
-
表 1 去噪实验对比结果
Table 1. Comparison results of denoising experiments
方法 X Y Z R/dB e R/dB e R/dB e DnCNN 82.142 1.563e-4 85.175 1.102e-4 82.234 1.547e-4 RED-CNN 83.495 1.338e-4 88.428 7.580e-5 85.440 1.070e-4 UNet 89.157 6.970e-5 89.257 6.890e-5 89.092 7.020e-5 CBDNet 92.837 4.562e-5 95.498 3.358e-5 95.235 3.462e-5 BRDNet 93.519 4.218e-5 96.901 2.858e-5 96.035 3.157e-5 PRIDNet 93.943 4.017e-5 96.284 3.068e-5 95.770 3.255e-5 3MNet 96.766 2.902e-5 97.257 2.743e-5 96.514 2.988e-5 表 2 3 MNet内部子网络实验对比结果
Table 2. 3 MNet internal sub-network experimental comparison results
方法 R/dB e(×10-5) Remove both 93.363 4.294 Remove image preprocessing 95.322 3.427 Remove multiscale convolution 95.647 3.301 3MNet 96.766 2.902 -
[1] REDLER G, EPEL B, HALPERN H J. Principal component analysis enhances SNR for dynamic electron paramagnetic resonance oxygen imaging of cycling hypoxia in vivo[J]. Magnetic Resonance in Medicine, 2013, 71(1): 440−450. [2] CHEN J, CHEN Q, LIANG C, et al. Albumin-templated biomineralizing growth of composite nanoparticles as smart nanotheranostics for enhanced radiotherapy of tumors[J]. Nanoscale, 2017, 146(39): 14826−14835. [3] EPEL B, BOWMAN M K, MAILER C, et al. Absolute oxygen R-1 e imaging in vivo with pulse electron paramagnetic resonance[J]. Magnetic Resonance in Medicine, 2014, 72(2): 362−368. doi: 10.1002/mrm.24926 [4] QIAO Z, REDLER G, EPEL B, et al. Comparison of parabolic filtration methods for 3D filtered back projection in pulsed EPR imaging[J]. Journal of Magnetic Resonance, 2014, 248: 42−53. doi: 10.1016/j.jmr.2014.08.010 [5] EPEL B, SUNDRAMOORTHY S V, BARTH E D, et al. Comparison of 250 MHz electron spin echo and continuous wave oxygen EPR imaging methods for in vivo applications[J]. Medical Physics, 2011, 38(4): 2045−2052. doi: 10.1118/1.3555297 [6] SIDKY E Y, KAO C M, PAN X. Accurate image reconstruction from few-views and limited-angle data in divergent-beam CT[J]. Journal of X-ray Science and Technology, 2006, 14(2): 119−139. [7] SIDKY E Y, PAN X. Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization[J]. Physics in Medicine and Biology, 2008, 53(17): 4777−4807. doi: 10.1088/0031-9155/53/17/021 [8] LIU Y, MA J, FAN Y, et al. Adaptive-weighted total variation minimization for sparse data toward low-dose X-ray computed tomography image reconstruction[J]. Physics in Medicine and Biology, 2012, 57(23): 7923−7956. doi: 10.1088/0031-9155/57/23/7923 [9] ZHANG Y, ZHANG W H, CHEN H, et al. Few-view image reconstruction combining total variation and a high-order norm[J]. International Journal of Imaging Systems and Technology, 2013, 23(3): 249−255. doi: 10.1002/ima.22058 [10] 闫慧文, 乔志伟. 基于ASD-POCS框架的高阶TpV图像重建算法[J]. CT理论与应用研究, 2021,30(3): 279−289. DOI: 10.15953/j.1004-4140.2021.30.03.01.YAN H W, QIAO Z W. High order TpV image reconstruction algorithms based on ASD-POCS framework[J]. CT Theory and Applications, 2021, 30(3): 279−289. DOI: 10.15953/j.1004-4140.2021.30.03.01. (in Chinese). [11] SIDKY E Y, REISER I, NISHIKAWA R M, et al. Practical iterative image reconstruction in digital breast tomosynthesis by non-convex TpV optimization[J]. Medical Imaging 2008: Physics of Medical Imaging, 2008: 6913. DOI: 10.1117/12.772796. [12] JAIN V, SEUNG H S. Natural image denoising with convolutional networks[C]//International Conference on Neural Information Processing Systems, 2008: 769-776. [13] MAO X, SHEN C, YANG Y B. Image restoration using very deep convolutional encoder-decoder networks with symmetric skip connections[J]. Advances in Neural Information Processing Systems, 2016, 29: 2802−2810. [14] ZHANG K, ZUO W, CHEN Y, et al. Beyond a gaussian denoiser: Residual learning of deep CNN for image denoising[C]//IEEE Transactions on Image Processing, 2017, 26(7): 3142-3155. [15] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778. [16] IOFFE S, SZEGEDY C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[J]. JMLRorg, 2015, 37: 448−456. [17] CHEN H, ZHANG Y, KALRA M K, et al. Low-dose CT with a residual encoder-decoder convolutional neural network[J]. IEEE Transactions on Medical Imaging, 2017, 36(12): 2524−2535. doi: 10.1109/TMI.2017.2715284 [18] ZHANG K, ZUO W, ZHANG L. FFDNet: Toward a fast and flexible solution for CNN based image denoising[J]. IEEE Transactions on Image Processing, 2018, 27(9): 4608−4622. doi: 10.1109/TIP.2018.2839891 [19] 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, 2015: 234-241. [20] GUO S, YAN Z, ZHANG K, et al. Toward convolutional blind denoising of real photographs[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 1712-1722. [21] TIAN C, XU Y, ZUO W. Image denoising using deep CNN with batch renormalization[C]//Neural Networks, 2019, 121: 461-473. [22] YU F, KOLTUN V. Multi-scale context aggregation by dilated convolutions[C]//Intermational Conference on Learning Representations, 2016. [23] ZHAO Y Y, JIANG Z Q, MEN A, et al. Pyramid real image denoising network[C]//2019 IEEE Visual Communications and Image Processing (VCIP), 2020. [24] MILDENHALL B, BARRON J T, CHEN J, et al. Burst denoising with kernel prediction networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 2502-2510. [25] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7132-7141. [26] WOO S, PARK J, LEE J Y, et al. Cbam: Convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision (ECCV), 2018: 3-19. [27] SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-v4, inception-resnet and the impact of residual connections on learning[C]//Thirty-First AAAI Conference on Artificial Intelligence. San Francisco: 2017, 31(1): 4278-4284. [28] WU Z, SHEN C, Van den HENGEL A. Wider or deeper: Revisiting the resnet model for visual recognition[J]. Pattern Recognition, 2019, 90: 119−133. doi: 10.1016/j.patcog.2019.01.006 [29] ZHANG Y, TIAN Y, KONG Y, et al. Residual dense network for image super-resolution[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 2472-2481. [30] LIU C, SHANG Z, QIN A. A multiscale image denoising algorithm based on dilated residual convolution network[C]//Chinese Conference on Image and Graphics Technologies, 2019: 193-203. [31] WANG T, SUN M, HU K. Dilated deep residual network for image denoising[C]//2017 IEEE 29th International Conference on Tools with Artificial Intelligence (ICTAI), 2017: 1272-1279. [32] LU Y, JIANG Z, JU G, et al. Recursive multi-stage upscaling network with discriminative fusion for super-resolution[C]//2019 IEEE International Conference on Multimedia and Expo (ICME), 2019: 574-579. [33] 陈康, 狄贵东, 张佳佳, 等. 基于改进U-net卷积神经网络的储层预测[J]. CT理论与应用研究, 2021,30(4): 403−415. DOI: 10.15953/j.1004-4140.2021.30.04.01.CHEN K, DI G D, ZHANG J J, et al. Reservoir prediction based on improved U-net convolutional Neural network[J]. CT Theory and Applications, 2021, 30(4): 403−415. DOI: 10.15953/j.1004-4140.2021.30.04.01. (in Chinese). -