Article Preview
TopIntroduction
Non-invasive diagnosis techniques have always been a preferred approach in the investigation of any disease or disorder in the body. Magnetic resonance imaging is an excellent non-invasive diagnosis technique which makes use of nuclear magnetic resonance for providing internal images of the body so that any pathological alterations of the body’s living tissues can be clearly discovered. MR imaging works by first placing the desired object in powerful magnetic field to excite the hydrogen nuclei present in water and lipids and then studying the hydrogen nuclei’s relaxation properties. The atomic nuclei spin then arrange themselves in some manner with respect to the applied magnetic field. To obtain different images of the object under consideration, magnetic gradients in X, Y and Z axis are applied. These magnetic gradients are orthogonal in nature. A 2D or 3D matrix created from this data helps in the formation of the scanned image of the body part or soft tissue or organ. The best part of MR imaging is that it does not use any harmful ionizing radiation, which makes it a popular medical imaging technique. MR images are used in visualizing tumors, blood vessels, infections in body organs, strokes, degenerative spinal diseases, torn ligaments, shoulder injuries, etc. In the image acquisition process of MR images, various kind of noise corrupts the images. The noise is mainly thermal noise from the patient and some additional noise from hardware. Removing noise from MR images is an important task of biomedical imaging and has been greatly explored by researchers over the time. In denoising MR images, the fine structures present in the image should be secured as they aid in correctly diagnosing the MR scan.
There are various methods for noise removal from images. Some of the popular methods include neighborhood filters (Smith & Brady, 1997; Tomasi & Manduchi, 1998; Yaroslavsky,1985), total variation (TV) method (Rudin et al., 1992), non local means (NLM) and its advancements (Buades et al., 2005; Kervrann & Boulanger, 2006) and wavelet transformations (Donoho & Johnstone, 1994; Chang et al., 2000). Buades et al. (2005) introduced the NLM filter for denoising 2D images and it has undergone various advancements. Non local means filter has widely been used for denoising MR images. Coupe et al. (2008) proposed an optimised 3D blockwise NLM (ONLM) for MR images. It focused on selecting the appropriate voxel in the search volume. Blockwise implementation of NLM and also modified the smoothing parameter. A rician NLM filter for MR image was given by Daessle et al. (2008). This filter was further enhanced by using combined patch and pixel similarity (RNLM-CPP) and it also removed both Gaussian as well as rician noise (Zhang et al., 2014) from MR images. Manjon et al. (2009) suggested using principal component analysis in NLM filter to deal with multi-component MR images (MNLM). This MNLM method could be used with images with low noise and the use of principal component analysis helps in removing extra noise. Yaniv et al. (2010) presented a Dynamic NLM (DNLM) for dealing with the redundant information present in MR images. The NLM technique was merged with Discrete Cosine Transform (DCTNLM) by Hu et al. (2012) and it was used for denoising MR images. Vega et al. (2012) reduced the computation speed of NLM by examining the prominent properties of the pixels under consideration. Kang et al. (2013) suggested an adaptive NLM (ANLM) technique for noise removal from MR images and also preserved edges. Aksam et al. (2013) suggested some improvements in Adaptive NLM (IANLM) for removing noise from brain MRI by mixing wavelet coefficients and removing rician noise. Yang et.al (2015) proposed a pre-smooth NLM (PNLM) technique for removing noise from MRI by using image transformation. Joshi et al. (2016) reviewed various optimization techniques of NLM and further discussed a denoising approach using a combination of median filter, wiener filter and NLM filter for noise removal from MR images (2017). Table 1 gives a summary of different NLM based techniques used in denoising MRI.