Modified Anisotropic Diffusion Filtering Algorithm for MRI Aditya Srivastava Department of Electronics and Communication Engineering, SRMGPC, Lucknow-227105 (U.P), India. Email Id: adityasri1092@gmail.com Vikrant Bhateja Member, IEEE Department of Electronics and Communication Engineering, SRMGPC, Lucknow-227105 (U.P), India. Email Id:bhateja.vikrant@ieee.org Harshit Tiwari Department of Electronics and Communication Engineering, SRMGPC, Lucknow-227105 (U.P), India. Email Id:htiwari101092@gmail.com Abstract – During the acquisition process of Magnetic Resonance Imaging (MRI), irregular bias is imposed in the intensity values of the pixels. These biases follow the Gaussian Noise distribution model and act as a constraint to the effective medical diagnosis. The conventional Anisotropic Diffusion (AD) approach is limited to preserve the structural integrity of MRI at only low noise levels. This paper proposes a modified AD algorithm aimed to improve the estimation of the diffusion constant to facilitate better edge detection and preservation of details. The proposed algorithm operates on the decomposed mask images of MRI by incorporating the domain filtering principle of the Bilateral filter(prior to the estimation of diffusion constant). Simulation trials have been conducted at different Gaussian noise variances and performance has been evaluated on the basis of Peak Signal- Noise Ratio (PSNR) and Structural Similarity (SSIM). The proposed algorithm has shown stable value of evaluation parameters at higher noise variances. Also, the preservation of details has improved as compared to the conventional AD approach. Keywords – AD, Bilateral Filter, Gaussian Noise, SSIM. I. INTRODUCTION The recent years have seen the emergence of MRI as a powerful diagnostic technique for the detailed visualization of the internal structures of the human body. However, the image quality of the obtained MRI is constrained by the presence of large amount of noise introduced during the acquisition process [1]. This noise, usually modelled as Gaussian noise [2]-[3], introduces irregular intensity bias in the pixel values, thereby hampering the performance of the various image post- processing modules such as enhancement [4]-[8], segmentation [9]-[10] and classification [11]-[13]. Thus, removal of this unwanted noise becomes a fundamental process in the medical image processing. Over the years, various denoising approaches [14]-[18] based on linear smoothening [19] have been developed for the suppression of the noise. However, these approaches eliminated the noise at the expense of the fine details and tissue edges which were lost due to blurring. This lead to the development of edge preserving AD approach introduced by Perona and Malik [20] and was based on the scale space concept introduced by [21]. This approach overcame the limitations of the linear smoothening approaches, such as blurring and loss of details, by suppressing the noise while respecting the tissue edges and small structures present within the image. Gerig et al. in their work [22] utilized this technique for noise suppression in MRI. Further improvements in the denoising ability of the AD approach were carried out in the works of [23]-[24]. An analysis on the behavior of the denoising mechanism of AD was performed in the work of [25]. This was utilized by Black et al. for the development of a robust AD filter [26] which incorporated the robust statistics in the AD filter. A fourth order partial differential equation based denoising approach was developed by [27] which utilized the concept of signal dependent noise characteristics in MRI [28]. The application of AD filter was extended for spatially varying noise levels in MRI in the work carried out by Samsonov and Johnson [29]. Recently, an extension of the AD filter based on the estimation of noise level has been developed [30]. In this paper, a modification in the estimation of the diffusion constant of the AD filter has been proposed, where the decomposed mask images have been filtered using the domain function of the bilateral filter [31]-[32] as a pre-processing step. In the remaining part of the paper, Section II details the AD process and the proposed filtering algorithm, Section III presents the simulation results and their subsequent discussions and Section IV includes the conclusion part. II. PROPOSED METHODOLOGY A. Anisotropic Diffusion (AD) Filtering AD is a widely used filtering algorithm for the biomedical images [33]-[36], specifically MRI. It is a filtering technique aimed at the noise suppression without the removal of the significant parts of the image content, typically the edges and boundaries [26]. This technique incorporates the non-linear and space-variant transformation of the original MRI, so as to 1885 978-9-3805-4416-8/15/$31.00 c 2015 IEEE