摘要
针对低剂量CT(Computed Tomography)重建图像质量退化的问题,提出一种基于小波收缩和绝对差值排序各项异性扩散的MLEM(Maximum Likelihood Expectation Maximization)低剂量CT重建算法。算法在每次迭代中首先采用MLEM算法对低剂量CT投影数据进行重建。由于各项异性扩散对噪声敏感,所以算法先对重建后的图像进行小波变换,再在更稳定的低频小波域进行基于绝对差值排序的各项异性扩散处理,对小波高频系数进行软阈值降噪处理。然后将降噪处理后的系数进行小波反变换,得到降噪后的图像。最后使用中值滤波对图像进行处理,从而消除脉冲噪声点。实验结果表明,与其他几种常用重建算法相比,该算法重建的图像信噪比更高,归一化均方误差更小,处理后的图像更清晰,即可以在抑制噪声的同时,较好地保持图像的边缘和细节信息。
Concerning the problem of quality degradation of low-dose CT reconstruction images, we presented an MLEM low-dose CT reconstruction method which is based on wavelet shrinkage and rank-ordered absolute differences anisotropie diffusion. In each time of iteration, the algorithm first uses MLEM to reconstruct the low-dose projection data. Since the anisotropie diffusion is sensitive to noises, so the algorithm performs wavelet transform on the reconstructed image prior to conducting anisotropie diffusion processing based on rank-ordered absolute differences in more stable low-frequency wavelet domain and then carries out the soft threshold denoising processing on high- frequency coefficient of wavelet. After that the algorithm performs inverse discrete wavelet transform (IDWT) on the coefficients with denoising treatment and obtains the denoised images. Finally it uses median filter to process the image so as to eliminate the impulse noise points. Experimental results showed that compared with some other common reconstruction algorithms, the image reconstructed by the proposed one had higher signal-to-noise ratio (SNR) and smaller normalised mean square error, the processed image with the proposed algorithm were much clear, i.e. , it could well preserve edges and details information of image while suppressing the noise.
出处
《计算机应用与软件》
CSCD
2016年第2期156-160,178,共6页
Computer Applications and Software
基金
国家自然科学基金项目(6107119261271357
61171178)
山西省国际合作项目(2013081035)
山西省研究生优秀创新项目(2009011020-2)
山西省研究生优秀创新项目(20123098)
中北大学第十届研究生科技基金项目(20131035)
中北大学2013年校科学基金计划
山西省高等学校优秀青年学术带头人支持计划项目
关键词
低剂量CT
小波变换
图像重建
绝对差值排序
各向异性扩散
Low-dose computed tomography (CT) Wavelet transform Image reconstruction Rank-ordered absolute differences(ROAD) Anisotropic diffusion