Comparative analysis of noise removal techniques in MRI brain images

Abstract

Noise removal techniques have become an essential exercise in medical imaging applications, for the study of anatomical structures. To address this issue many denoising algorithm has been proposed both in spatial and frequency domain. Among them, few techniques in spatial domain are hybrid median filter, Weiner filter, bilateral filter, histogram equalization and in frequency domain are wavelet transform, independent component analysis were successfully used in medical imaging. The most commonly affected noises in medical image are salt and pepper, Gaussian, Speckle and Brownian noise. In this paper, the medical images taken for comparison include MRI brain images, in gray scale and RGB. The performances of these algorithms are analyzed for various noise types at different noise levels ranging from 0 dB to 30 dB. The evaluation of these algorithms is done by measures like peak signal to noise ratio (PSNR), root mean square error value (RMSE), universal quality index (UQI) and picture quality scale(PQS). Experimental results suggest that, independent component analysis performs better for removing salt and pepper noise in RGB and gray scale and Gaussian noise for images in RGB. Wavelet transform gives superior performance for removing speckle and Brownian noise for images in RGB and grayscale, irrespective of the noise level considered. Whereas histogram equalization gives better quality results while removing Gaussian noise at all noise levels for the images in gray scale only. On the other hand all spatial filtering techniques give comparative results at all dB levels in gray scale, which is inferior to frequency domain techniques.