小波域扩张网络用于低剂量CT图像快速重建

doi:10.19665/j.issn1001-2400.2020.04.012

摘要/Abstract

摘要：

低剂量计算机断层扫描技术具有低辐射和高效的优点,但低剂量图像伴随的噪声和伪影降低了诊断的可靠性。为了有效地改善低剂量图像的质量,通过借助小波域的优势来增强重建图像的细节纹理,结合改进的扩张卷积和亚像素技术来提高运算速度,使模型更好地部署到断层扫描设备。使用“2016 AAPM低剂量图像挑战赛”数据集对提出的方法进行评估。实验结果表明,该方法重建图像的视觉效果提升明显,平均峰值信噪比比RED-CNN约高0.1428dB(1mm)、0.0939dB(3mm),在中央处理器和图形处理器上的运算速度分别提升约55%、50%以上。

关键词: 图像重建, 计算机断层扫描, 小波变换, 卷积神经网络

Abstract:

Low-dose CT has the advantages of low radiation and high efficiency, but the noise and artifacts with low-dose CT images reduce the reliability of diagnosis. In order to improve the quality of low-dose CT images, this paper attempts to enhance the visuals of reconstructed CT images in the wavelet domain, and improve the running speed by combining the multi-dilated convolution and subpixel, so that the model can be better deployed to the CT equipment. The data set of "2016 AAPM Low Dose CT image Challenge" is used to evaluate the proposed method. Experimental results show that the visuals of reconstructed CT images are better. Compared with RED-CNN, the average PSNR of the proposed method is improved by 0.1428dB (1mm) / 0.0939dB (3mm), and the running speed on the CPU and GPU is increased by more than 55% and 50%, respectively.

Key words: image reconstruction, computed tomography, wavelet transforms, convolutional neural networks

中图分类号:

TP391

李坤伦,张鲁,许宏科,宋焕生. 小波域扩张网络用于低剂量CT图像快速重建[J]. 西安电子科技大学学报, 2020, 47(4): 86-93.

LI Kunlun,ZHANG Lu,XU Hongke,SONG Huansheng. Waveletdomain dilated network for fast low-dose CT image reconstruction[J]. Journal of Xidian University, 2020, 47(4): 86-93.

图/表 8

图1

图2

图3

表1

表2

图4

图5

表3

参考文献 16

[1]	BRENNER D J, ELLISTON C D, HALL E J, et al. Estimated Risks of Radiation-induced Fatal Cancer from Pediatric CT[J]. American Journal of Roentgenology, 2001,176(2):289-296. pmid: 11159059
[2]	CHEN Y, YIN X, SHI L, et al. Improving Abdomen Tumor Low-dose CT Images Using a Fast Dictionary Learning Based Processing[J]. Physics in Medicine and Biology, 2013,58(16):5803-5820. doi: 10.1088/0031-9155/58/16/5803 pmid: 23917704
[3]	HA S, MUELLER K. Low Dose CT Image Restoration Using a Database of Image Patches[J]. Physics in Medicine and Biology, 2015,60(2):869-882. pmid: 25565336
[4]	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 pmid: 28622671
[5]	WOLTERINK J M, LEINER T, VIERGEVER M A, et al. Generative Adversarial Networks for Noise Reduction in Low-dose CT[J]. IEEE Transactions on Medical Imaging, 2017,36(12):2536-2545. doi: 10.1109/TMI.2017.2708987 pmid: 28574346
[6]	MEHTA S, RASTEGARI M, CASPI A, et al. Espnet: Efficient Spatial Pyramid of Dilated Convolutions for Semantic Segmentation[C]// Lecture Notes in Computer Science: 11214. Heidelberg: Springer Verlag, 2018: 561-580.
[7]	PENG Y, ZHANG L, LIU S, et al. Dilated Residual Networks with Symmetric Skip Connection for Image Denoising[J]. Neurocomputing, 2019,345:67-76. doi: 10.1016/j.neucom.2018.12.075
[8]	LU Z, YU Z, YA-LI P, et al. Fast Single Image Super-resolution via Dilated Residual Networks[J]. IEEE Access, 2018,7:109729-109738. doi: 10.1109/Access.6287639
[9]	GUO S, YAN Z, ZHANG K, et al. Toward Convolutional Blind Denoising of Real Photographs[C]// Proceedings of the 2019 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2019: 1712-1722.
[10]	欧阳宁, 王先傲, 蔡晓东, 等. 结合小波与递归神经网络的低分辨率人脸识别[J]. 西安电子科技大学学报, 2019,46(6):95-101.
	OUYANG Ning, WANG Xian’ao, CAI Xiaodong, et al. Low Resolution Face Recognition Method Based on Wavelet and Recursive Neural Networks[J]. Journal of Xidian University, 2019,46(6):95-101.
[11]	刘树东, 王晓敏, 张艳. 一种对称残差CNN的图像超分辨率重建方法[J]. 西安电子科技大学学报, 2019,46(5):15-23.
	LIU Shudong, WANG Xiaomin, ZHANG Yan. SymmetricResidual Convolution Neural Networks for the Image Super-resolution Reconstruction[J]. Journal of Xidian University, 2019,46(5):15-23.
[12]	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
[13]	WANG P, CHEN P, YUAN Y, et al. Understanding Convolution for Semantic Segmentation[C]// Proceedings of the 2018 IEEE Winter Conference on Applications of Computer Vision. Piscataway: IEEE, 2018: 1451-1460.
[14]	SHI W, CABALLERO J, HUSZAR F, et al. Real-time Single Image and Video Super-resolution Using an Efficient Sub-pixel Convolutional Neural Network[C]// Proceedings of the 2016 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2016: 1874-1883.
[15]	MCCOLLOUGH C H, BARTLEY A C, CARTER R E, et al. Low‐dose CT for the Detection and Classification of Metastatic Liver Lesions: Results of the 2016 Low Dose CT Grand Challenge[J]. Medical Physics, 2017,44(10):339-352.
[16]	FLOHR T G, STIERSTORFER K, ULZHEIMER S, et al. Image Reconstruction and Image Quality Evaluation for a 64-slice CT Scanner with Z-flying Focal Spot[J]. Medical Physics, 2005,32(8):2536-2547. doi: 10.1118/1.1949787 pmid: 16193784

网络结构	平均峰值信噪比/dB	平均测试时间/s
DWT + CNN+ IWT	24.7738	1.4430 / 0.1856
DWT +CNN + Subpixel	24.7763	1.4235 / 0.1749

条件	性能	LDCT	CNN10	RED-CNN	FFWDNet	FWDNet
在1mm的对比	平均RMSE	45.7471	23.5903	23.3285	23.2580	23.1206
	平均PSNR/dB	18.8504	24.6014	24.6336	24.7242	24.7763
	平均SSIM	0.6125	0.6621	0.6671	0.6713	0.6725
在3mm的对比	平均RMSE	28.6170	14.9904	15.0339	14.9627	14.8723
	平均PSNR/dB	22.9338	28.5423	28.5173	28.5589	28.6112
	平均SSIM	0.6732	0.7441	0.7425	0.7459	0.7487

运算时间	设备	CNN10	RED-CNN	FFWDNet	FWDNet
平均测试时间/s	CPU	51.36 / 2.568	91.40 / 4.570	39.00 / 1.950	39.24 / 1.962
平均测试时间/s	GPU	4.46 / 0.223	6.00 / 0.300	2.88 / 0.144	3.70 / 0.185
平均训练时间/h	GPU	12	14	10	10