An Enhanced Pseudocolor-Guided Non-Local Means Approach for Alzheimer’s MRI Denoising

Alzheimer’s Disease (AD) is a progressive neurodegenerative disease. Magnetic resonance imaging (MRI) is frequently used to investigate brain structure in AD research. MRI images may be corrupted by noise and preprocessing is important for noise suppression while preserving anatomical details and low contrast tissue variations. Non-Local Means (NLM) denoises images by averaging similar patches. Patch similarity based on greyscale may not be sufficient to distinguish subtle patterns in low-contrast regions. In this paper, we propose a pseudocolor-guided Non-Local Means (PC-NLM) framework for brain MRI denoising. The similarity weights are computed by combining the greyscale and pseudocolor patch distances with a convex combination. The denoised intensities are reconstructed from the original greyscale image. The method was validated on sagittal slice 66 from 50 subjects of the Open Access Series of Imaging Studies (OASIS) dataset with synthetic Gaussian noise of variance 0.01, 0.02 and 0.03. PC-NLM yielded a higher peak signal-to-noise ratio (PSNR) than conventional NLM (22.244 dB vs 21.773 dB), a higher structural similarity index (SSIM; 0.253 vs 0.236) and a lower mean squared error (MSE; 0.005965 vs 0.006648) for a noise variance of 0.03. PC-NLM achieved the highest PSNR and MSE for the reported noise levels among the methods considered, while median filtering achieved the highest SSIM. The results show that pseudocolor guidance can improve the performance of denoising by PSNR and MSE compared to conventional Non-Local Means (NLM) under the tested conditions, while median filtering can achieve a higher SSIM.