Investigational Analysis of Denoising Filters on Brain MR Images using Image Quality Assessment Performance Measurements

Abstract

Magnetic resonance imaging (MRI) is one of the most widely used imaging techniques in medical imaging with better capability for soft tissues imaging. It is very appropriate for brain imaging, for early detection and diagnosis of brain tumors, treatment monitoring and, other brain abnormalities. Even though it’s providing high-quality information, images may be corrupted by the different types of noises during acquisition and transmission. However, the incorporated noise during MRI image acquisition and transmission process makes it difficult for human interpretation as well as computer-aided analysis of the images. The Magnetic Resonance Imaging (MRI) images are generally prone to noises such as Gaussian noise, Rician noise, Salt and Pepper noise (impulse noise) and speckle noise. A more precise brain MR image is essential for further analysis process. In this proposed experimentation five different filters such as Median filter, Adaptive Median filter, Gaussian filter, Wiener filter and Average (mean) filter are used to remove the additive noises present in the MRI images. The additive noises used to test filters are Gaussian, Salt and pepper, Rician and speckle noises. The additive noise of varying density was step by step added to MRI image from 10% to 90% to evaluate performance of the denoising filters by qualitative and quantitative evaluation. The performances analyses of these denoising filtering techniques are evaluated by observing the statistical parameters such as Peak Signal-to-Noise Ratio (PSNR), Signal-to-Noise Ratio (SNR), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Normalized Mean Squared Error (NRMSE). The proposed experimental investigation shows that Median filter reconstructs a high quality image from the noised image than other filters in Salt and pepper, Rician, Salt and Pepper with PSNR=21.64, MSE=444, RMSE=21.09, NRMSE=0.08, SNR=5.09 at 10% noise variance and Rician with PSNR=31.84, MSE=42.56, RMSE=6.52, NRMSE=0.02, SNR=12.19 at 10% noise variance. While for Gaussian noise removal, Gaussian filter outperformed other filters with PSNR=32.42(db), MSE= 37.20, RMSE =6.09, NRMSE =0.02, SNR=8.06 at 10% noise variance. Whereas for speckle noise removal, Weiner filter performs better than others which result of PSNR=32.65, MSE=35.32, RMSE=5.94, NRMSE=0.02, SNR=7.98 at 10% noise variance. This experimental investigation shows that the Median, Gaussian and Weiner filter produces better denoising effects, by preserving the main structures and details.