適用於核磁共振影像的非局部平均雜訊消除演算法

Abstract

當醫生在為病患進行診斷時，常常需要各式各樣的醫學儀器來輔助，以增加其診斷的正確性。其中，醫學影像就是一種常被廣泛使用的輔助工具，例如：核磁共振影像(Magnetic Resonance Imaging)、超音波影像(Ultrasound Imaging)、電腦斷層掃描影像(Computed Tomography) 、正子放射型斷層掃描影像(Positron Emission Tomography)。然而，在取得醫學影像的過程中，常常伴隨著隨機雜訊的干擾，嚴重影響影像的的品質，使得後續的醫學影像分析工作困難重重，例如：切割(segmentation)、套合(registration)。因此，醫學影像的除雜訊是進行醫學影像分析工作中一項非常重要的前置作業。 非局部區域平均(Non-local Means, NLM)演算法，是近幾年來所提出非常有效的一般數位影像除雜訊方法。在此論文中，我們嘗試藉由改善非局部平均演算法，並利用主成分分析(Principal Component Analysis，PCA)為輔，提出適用於去除核磁共振影像的雷斯雜訊(Rician Noise)之有效方法。不僅可以大量降低運算複雜度，且在提升除雜訊之影像品質的同時，亦可保留更多影像細節部分。

There is a variety range of medical diagnostic techniques to assist doctor in diagnosis. In fact, medical image which is often composed of low-contrast objects corrupted with random noise arising in the image acquisition process is widely used during the diagnosis of a doctor. In order to remove noise from digital medical image to improve visual quality and performance, digital medical image denoising has attached increasing attention and become a fundamental task in medical image analysis. The recently proposed non-local means algorithm is a classical noise removal method in digital image, and the algorithm really present excellent results. Non-local means algorithm takes the advantage of the nature redundancy information in images replaces the gray value of noisy image with the average of the pixel in non-local areas. Owing to the effective denoising ability of non-local means filter, we do our utmost to make some regulations to improve the ability of the denoising scheme. In this work, we try to preserve the edge of object and speed up the algorithm while removing Rician noise from MRI.