Noise Removal from Brain MRI Images Using Adaptive Bayesian Shrinkage

Abstract

In MR machine, surface coils, especially phased-arrays are used extensively for acquiring MR images with high spatial resolution. The signal intensities on images acquired using these coils have a non-uniform map due to coil sensitivity profile. Although these smooth intensity variations have little impact on visual diagnosis, they become critical issues when quantitative information is needed from the images. Sometimes, medical images are captured by low signal to noise ratio (SNR). The low SNR makes it difficult to detect anatomical structures because tissue characterization fails on those images. Hence, denoising are essential processes before further processing or analysis will be conducted. They found that the noise in MR image is of Rician distribution. Hence, general filters cannot be used to remove these types of noises. The linear spatial filtering technique blurs the object boundaries and degrades the sharp details. The existing works proved that Wavelet based works eliminates the noise coefficient that called wavelet thresholding. Wavelet thresholding estimates the noise level from high frequency content and estimates the threshold value by comparing the estimated noisy wavelet coefficient with other wavelet coefficients and eliminate the noisy pixel intensity value. Bayesian Shrinkage rule is one of the widely used methods. It uses for Gaussian type of noise, the proposed method introduced some adaptive technique in Bayesian Shrinkage method to remove Rician type of noises from MRI images. The results were verified using quantitative parameters such as Peak Signal to Noise Ratio (PSNR). The proposed Adaptive Bayesian Shrinkage Method (ABSM) outperformed existing methods.