Fundamental Filters Performances Based on Window Sizes for T2-Weighted MRI Images

In magnetic resonance imaging (MRI) images, noise is a common issue and it is locally signal dependent. Theoretical expectation of noise measured in unfiltered images is found to be normally distributed, spatially invariant and white. Therefore, the noise has to be removed before processing stage. In this work, there are three different filtering algorithms which are median filter (MF), adaptive filter (ADF) and average filter (AVF) used. These filters removed additive noises which are Gaussian, Salt and pepper and speckle noises based on different window sizes of 3x3 and 5x5 which were present in the MRI images. The noise density was gradually added to the MRI image to up to its 90% to evaluate the performance of the filters qualitatively and quantitatively. They were later compared by employing the statistical parameters such as mean squared error (MSE) and peak signal-to-noise ratio (PSNR). The study concludes that the mask size of the filter has significant impact on the PSNR and MSE. The 3x3 size MF showed its capability to reconstruct a higher quality image and perform better during the speckle noise removal than the other filters. The MF also produced better de-noising results and preserving the main structures and details for 3x3 window size. KeywordsDenoising; filtering; window sizes; MRI image; image pre-processing