医学CT序列图像的混合去噪算法

doi:10.12068/j.issn.1005-3026.2024.04.002

摘要/Abstract

摘要：

医学CT（computer tomography）序列图像会因为各种原因而掺杂噪声，去噪能够有效地提高图像的质量.常见的去噪算法都是针对单张图像进行，考虑到CT序列图像之间具有很高的相似性，本文提出一种基于相邻图像结构相似性的混合去噪算法.该算法首先计算序列图像的最大和最小灰度值，根据灰度值绘制直方图，设定相关阈值参数，根据筛选之后的直方图计算窗宽窗位，然后进行调窗处理.之后计算目标图像与其前后相邻图像之间的结构相似性，最后根据结构相似性对3张图片混合使用BM3D和高斯滤波2种去噪算法.通过对比实验表明，该算法在均方误差、峰值信噪比和结构相似性三方面都有所提高，能够有效地提高图像质量.

关键词: 医学CT序列图像, 结构相似性, 混合去噪算法, 直方图, 调窗处理

Abstract:

The medical CT sequence images dopes noise for various reasons. Denoising can effectively improve image quality. The common algorithms are used for single image， while the CT sequence images have high similarity between adjacent images. Therefore， this paper proposes a hybrid denoising algorithm based on the structural similarity. Firstly， a histogram is drawn according to the maximum and minimum gray value. Secondly， relevant threshold parameters are set to calculate the window width and window level， and then conduct window adjustment. Thirdly， the structural similarity of the target image and its adjacent images are calculated. Finally， BM3D and Gaussian filtering algorithms are mixed for three images according to structural similarity. Experimental results show that the algorithm can improve the mean square error， peak signal?to?noise ratio and structural similarity， which effectively improves the image quality.

Key words: medical CT sequence images, structural similarity, hybrid denoising algorithm, histogram, window adjustment

中图分类号:

TP 391.41

陈锦林, 原培新. 医学CT序列图像的混合去噪算法[J]. 东北大学学报（自然科学版）, 2024, 45(4): 464-473.

Jin-lin CHEN, Pei-xin YUAN. Hybrid Denoising Algorithm for Medical CT Sequence Images[J]. Journal of Northeastern University(Natural Science), 2024, 45(4): 464-473.

图/表 19

参考文献 22

1	Balda M， Hornegger J， Heismann B.Ray contribution masks for structure adaptive sinogram filtering［J］.IEEE Transactions on Medical Imaging，2012，31（6）：1228-1239.
2	Chen H， Zhang Y， Zhang W H，et al.Low‑dose CT via convolutional neural network［J］.Biomedical Optics Express，2017，8（2）：679-694.
3	Nishio M， Nagashima C， Hirabayashi S，et al.Convolutional auto‐encoder for image denoising of ultra‐low‐dose CT［J］.Heliyon，2017，3（8）：e00393.
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.
5	Yamanappa W， Sudeep P V， Sabu M K，et al.Non‐local means image denoising using Shapiro‑Wilk similarity measure［J］.IEEE Access，2018，6：66914-66922.
6	Dabov K， Foi A， Katkovnik V，et al.Image denoising by sparse 3‐D transform‐domain collaborative filtering［J］.IEEE Transaction on Image Processing，2007，16（8）：2080-2095.
7	Elad M， Aharon M.Image denoising via sparse and redundant representations over learned dictionaries［J］.IEEE Transactions on Image Processing，2006，15（12）：3766-3745.
8	Aharon M， Elad M， Bruckstein A.K‐SVD：an algorithm for designing overcomplete dictionaries for sparse representation［J］.IEEE Transactions on Signal Processing，2006，54（11）：4311-4322.
9	Zoran D， Weiss Y.From learning models of natural image patches to whole image restoration［C］//International Conference on Computer Vison（ICCV）.Barcelona：IEEE，2011：479-486.
10	Huang S S， Zhu J B.Recovery of sparse signals using OMP and its variants：convergence analysis based on RIP［J］.Inverse Problems，2011，27（3）：e035003.
11	Ohhashi A， Yamada S， Haruki K，et al. Automatic adjustment of display window （gray level） for MR images using a neural network［C］//International Society for Optics and Photonics. San Jose：SPIE，1991：63-74.
12	周振环，陈思平，陶笃纯，等.医学图像的自动调窗与分割［J］.生物医学工程学杂志，2005，22（2）：331-334.
	Zhou Zhen‐huan， Chen Si‐ping， Tao Du‐chun，et al.Medical image automatic adjusting window and segmentation［J］.Journal of Biomedical Engineering，2005，22（2）：331-334.
13	沈琴，蒋谟文，骆建华.一种改进的磁共振图像自动调窗算法［J］.中国医疗器械杂志，2011，35（4）：253-255.
	Shen Qin， Jiang Mo‐wen， Luo Jian‐hua.An improved auto‐window algorithm for MR image［J］.Chinese Journal of Medical Instrumentation，2011，35（4）：253-255.
14	吕磊，赵勋杰.一种16位灰度图像自动调窗算法［J］.光电技术应用，2016，31（4）：27-30，45.
	Lei Lyu， Zhao Xun‐jie.Auto‐window algorithm for 16‐bit grayscale image［J］.Electro‐Optic Technology Application，2016，31（4）：27-30，45.
15	Mantri M， Taran S， Sunder G.DICOM integration libraries for medical image interoperability：a technical review［J］.IEEE Reviews in Biomedical Engineering，2022，15：247-259.
16	Li X R， Morgan P S， Ashburner J，et al.The first step for neuroimaging data analysis：DICOM to NIfTI conversion［J］.Journal of Neuroscience Methods，2016，264：47-56.
17	Gueld M O， Kohnen M， Keysers D，et al.Quality of DICOM header information for image categorization ［C］//Medical Imaging 2002：PACS and Integrated Medical Information Systems：Design and Evaluation. San Diego，California，2002：280-287.
18	Chen L， Li C Q， Li C.Security measurement of a medical communication scheme based on chaos and DNA coding［J］.Journal of Visual Communication and Image Representation，2022，83：103424.
19	Bashar F， El‐Sakka M R.BM3D image denoising using learning‐based adaptive hard thresholding［C］//Proceedings of the 11th Joint Conference on Computer Vision，Imaging and Computer Graphics Theory and Applications.Rome：IEEE，2015：204-214.
20	Elad M， Aharon M.Image denoising via sparse and redundant representations over learned dictionaries［J］.IEEE Transactions on Image Processing，2006，15（12）：3736-3745.
21	Sendur L， Selesnick I W.Bivariate shrinkage functions for wavelet‑based denoising exploiting interscale dependency［J］.IEEE Transactions on Signal Processing，2002，50（11）：2744-2756.
22	Vo B N， Vo B T， Hoang H G.An efficient implementation of the generalized labeled multi‐Bernoulli filter［J］.IEEE Transactions on Signal Processing，2017，65（8）：1975-1987.

评价标准	本文算法	维纳滤波	K-SVD	NLM	BM3D	中值滤波	均值滤波	高斯滤波	双边滤波	各向异性滤波
MSE	19.661 7	104.491 8	106.853 1	104.768 4	103.511 6	104.627 9	104.655 4	104.876 5	103.347 5	104.231 8
PSNR/dB	35.685 9	27.974 3	27.871 4	27.961 2	28.017 8	27.968 1	27.966 0	27.956 2	28.025 4	27.985 6
sim	0.777 4	0.574 7	0.660 0	0.594 5	0.554 0	0.581 2	0.597 0	0.605 9	0.547 4	0.575 3

评价标准	本文算法	维纳滤波	K-SVD	NLM	BM3D	中值滤波	均值滤波	高斯滤波	双边滤波	各向异性滤波
MSE	19.661 7	104.491 8	106.853 1	104.768 4	103.511 6	104.627 9	104.655 4	104.876 5	103.347 5	104.231 8
PSNR/dB	35.685 9	27.974 3	27.871 4	27.961 2	28.017 8	27.968 1	27.966 0	27.956 2	28.025 4	27.985 6
sim	0.777 4	0.574 7	0.660 0	0.594 5	0.554 0	0.581 2	0.597 0	0.605 9	0.547 4	0.575 3

评价标准	本文算法	维纳滤波	K-SVD	NLM	BM3D	中值滤波	均值滤波	高斯滤波	双边滤波	各向异性滤波
MSE	19.661 7	27.723 3	25.165 5	27.733 0	28.008 0	27.664 8	27.801 6	27.650 3	28.260 2	27.898 7
PSNR/dB	35.685 9	33.983 6	34.520 5	33.994 4	33.928 9	33.997 2	33.991 3	34.027 9	33.887 1	33.960 0
sim	0.777 4	0.749 6	0.812 8	0.769 5	0.743 0	0.756 1	0.771 8	0.780 4	0.724 4	0.753 6

评价标准	本文算法	维纳滤波	K-SVD	NLM	BM3D	中值滤波	均值滤波	高斯滤波	双边滤波	各向异性滤波
MSE	19.661 7	27.723 3	25.165 5	27.733 0	28.008 0	27.664 8	27.801 6	27.650 3	28.260 2	27.898 7
PSNR/dB	35.685 9	33.983 6	34.520 5	33.994 4	33.928 9	33.997 2	33.991 3	34.027 9	33.887 1	33.960 0
sim	0.777 4	0.749 6	0.812 8	0.769 5	0.743 0	0.756 1	0.771 8	0.780 4	0.724 4	0.753 6

部位	MSE
部位	本文算法	维纳滤波	K-SVD	NLM	BM3D	中值滤波	均值滤波	高斯滤波	双边滤波	各向异性滤波
骨架	31.436 0	57.095 9	55.454 0	56.807 2	56.748 1	57.113 0	54.555 6	54.158 4	56.710 8	56.838 4
膝盖	21.783 0	30.068 1	29.906 4	30.030 5	30.015 7	30.056 7	29.891 7	29.891 3	30.010 3	30.054 7
头部女	13.113 2	28.145 4	27.435 6	27.973 5	28.029 4	28.090 8	27.602 4	27.549 7	28.018 2	28.074 8
头部男	21.928 2	38.183 4	37.271 5	38.025 6	38.049 1	38.142 8	37.647 3	37.591 1	38.036 0	38.106 9
髋部女	55.292 5	74.290 9	73.923 5	74.235 7	74.213 6	74.270 9	74.089 4	74.077 8	74.207 0	74.271 1
髋部男	44.590 3	80.712 8	80.360 5	80.668 9	80.694 5	80.697 4	80.551 9	80.508 5	80.693 5	80.699 7
骨盆男	57.087 6	94.132 4	93.540 0	94.002 8	94.105 0	94.075 4	93.689 3	93.658 9	94.102 9	94.123 1
踝关节	24.592 6	29.602 8	29.259 8	29.549 5	29.564 5	29.587 7	29.414 7	29.413 5	29.563 1	29.586 5
肩部男	53.722 2	94.426 3	94.047 4	94.270 5	93.863 3	94.439 7	94.171 5	94.203 0	93.841 9	94.234 5