دانلود رایگان ترجمه مقاله حذف نویز از تصاویر DT-MR با PCA تکراری (نشریه الزویر 2015)

حذف نویز از تصاویر DT-MR با PCA تکراری

Denoising Of DT-MR Images With An Iterative PCA

Abstract

Nowadays most of the clinical applications uses Magnetic Resonance Images(MRI) for diagnosing neurological abnormalities. During MR image acquisition the emitted energy is converted to image by using some mathematical models, and this may cause addition of noise. Therefore we need to denoise the image. Currently most of the clinical application uses Diffusion Tensor-MR Images for tracking neural fibres by extracting features from the images. Noise in DT-MR Images make fibre tracking and disease diagnosing tougher. So our work aims to denoise the Diffusion Tensor MR images with better visual quality. In this paper, we propose a denoising technique that uses Structural Similarity Index Matrix (SSIM) for grouping similar patches and performs Iterative Principal Component Analysis on each group. By performing the weighted average on Principal Component, we have obtained the denoised DT-MR Image. For getting better visual quality of the denoised images we employ Iterative Principal component Analysis technique.

1 Introduction

Diffusion MRI (DMRI) is a magnetic resonance imaging (MRI) method which came into existence in the mid1980s. It allows the mapping of the diffusion process of molecules, mainly water, in biological tissues, in vivo and non-invasively. The Diffusion MRI, is also referred to as Diffusion Tensor Imaging or DTI has been extraordinarily successful in the field of medical science. It is mainly used for the study and treatment of neurological disorders, mostly for the patients with acute stroke. Because it can find the abnormalities in white matter fiber structure and provide models of brain connectivity. With Diffusion Tensor Imaging (DTI), diffusion anisotropy effects can be fully extracted, characterized, and exploited, providing even more exquisite details on tissue microstructure. DTI is used to demonstrate subtle abnormalities in a variety of diseases (including stroke, schizophrenia), for finding the issues related to fibre connectivity and is currently becoming part of many routine clinical protocols. The image acquisition process of DT-MR Images is by the patient is positioned within an MRI scanner where it forms a strong magnetic field around the area to be imaged. The medical applications rely on detecting a radio frequency signal emitted by excited hydrogen atoms in the body using energy from an oscillating magnetic field applied at the appropriate resonant frequency. The orientation of the image is controlled by varying the main magnetic field using gradient coils. As these coils are rapidly switched on and off they create the characteristic repetitive noises of an MRI scan. The contrast between different tissues is determined by the rate at which excited atoms return to the equilibrium state. During the tissue movement to equilibrium state and the anisotropic movement of water molecules, generated DTMRI may contain unwanted intensity values. This unwanted intensity values are called as Noise. The presence of noise makes disease diagnosing tougher. Therefore we need to denoise the image. The existing denoising algorithms may provide a denoised image but it may not yield good quality . ie, sometimes the denoised image may got blurred and may loss the fine structure. So our proposed Iterative PCA based method aims to denoise the DT-MRI which will result in good quality images. In this paper we propose a technique to perform denoising which is able to remove the noise with better quality thereby making the disease diagnosing easier. The denoising technique exploits the structural similarity of the patches and uses an iterative principal component analysis for the denoising process. The method takes advantage of the fact that there is a high degree of redundancy in the content of images. The proposed Iterative PCA based denoising technique is composed of four steps. In the first step, the similar patches are extracted and grouped by using Structural Similarity Index Matrix (SSIM) 4 and there is one-toone correspondence between patches and groups (sparse denoising). The actual denoising is done in the second step ie, by finding the principal component of each group. In the third step, patch restoration is performed by using the principal component. After the third step a denoised image will be obtained, but the first iteration helps us to remove only the most significant error portion of the image. The output of the first iteration needs to be refined for a better quality image. The amount of noise present in the output of first iteration is estimated and use this noise variance along with the denoised output of the first iteration is fed as the input to the second iteration. Iteration need to be stopped in the second stage, otherwise it may cause the loss of fine structures.