Research On Magnetic Resonance Image Denoising

This paper mainly studies the magnetic resonance imaging (MRI) denoising problem. In modern medicine, magnetic resonance imaging has become an important auxiliary tools for doctor to diagnosis and treatment of disease.There is a very important role of the denoising research in medicine. Most existing denoising method assumes that the noise variance is known, get rid of noise from the image directly. In fact, we don’t know the real image noise variance. This article will be to estimate the noise figure of the noise field, and then on the basis of noise estimation to build MRI image denoising model. More effective methods to eliminate noise are using existing image self-similarity in space domain, when the noise image with low SNR, the destruction of the self-similarity will reduce the quality of image restoration. At the same time, the original MRI image denoising model often does not take into account different regions of the denoising effect, the influence of the restored image is difficult to meet the needs of clinical application. In order to solve these problems, this paper presents a mixed non-convex denoising model. After model being established the quality of recovery also need to be considered. Medical images applied to clinical trials directly,it can provide comprehensive visual information for doctors and increase the accuracy of diagnosis. So in the clinical application of medical image quality put forward higher requirements, and the research of the image quality evaluation standard provided the foundation for practical for this requirement. Different from the traditional natural images, medical image has the characteristics of the diversity and nonstationarity, the traditional evaluation method for images may be not necessarily for MRI images, that find the appropriate MRI image quality evaluation criteria is also a key problem which need to solve. This paper mainly do the following three aspects:First, in this paper, we estimate the noise of the MRI image which noise field is in changed space. This article uses the Median absolute deviation(MAD) noise field in the wavelet domain to get the initial estimate, and then with the help of the correction factor to get the final changed space of noise variance, this chapter content will lay a foundation for the construction of subsequent MRI image denoising model.Second, based on the original MRI denoising models we construct a new MIR image model which named mixed non-convex denoising model. Combined with MRI images obey the characteristics of non-central chi-square, with maximum posteriori probability, we build a low SNR MRI image of mixed non-convex denoising model. In view of the different regions of the image by noise pollution degree of different characteristics, we build a detection operator that can automatically recognize image edge, texture and smooth area through the analysis of the local structure of image feature, which can improve the effect of image denoising.Third, in this paper, we adopt the appropriate medical image quality evaluation standards to measure the quality of recovery of the MRI image. Because the construction of SSIM objective evaluation standard is based on the human eye vision system, which can be highly adaptive to extract the structure information in the scene, on the basis of the theory of the adjustment of medical image quality. There is a great significance on adjusting the medical image quality and optimizing the medical image processing algorithm. In this article we will use the SSIM evaluation criteria, and connecting with the PSNR of the evaluation criteria to evaluate and feedback the MRI image denoising model together, which provides reliable image quality evaluation standards for clinical application and promotes accuracy for medical imaging diagnostic.