Performance evaluation of 3D median modified Wiener filter in brain T1-weighted magnetic resonance imaging

• Published in: Nuclear instruments & methods in physics research. Section A, Accelerators, spectrometers, detectors and associated equipment Vol. 1047; p. 167779
• Main Authors: Lee, Dohwa, Yun, Chang-Soo, Kang, Seong-Hyeon, Park, Minji, Lee, Youngjin
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.02.2023
• Subjects: 3D median modified wiener filter (MMWF); Alzheimer’s disease neuroimaging initiative (ADNI) database; BrainWeb simulated brain MRI database; Noise reduction filter; Quantitative evaluation of image quality; T1-weighted magnetic resonance brain image; BrainWeb simulated brain MRI database; Noise reduction filter; Alzheimer’s disease neuroimaging initiative (ADNI) database; Quantitative evaluation of image quality; 3D median modified wiener filter (MMWF); T1-weighted magnetic resonance brain image
• ISSN: 0168-9002; 1872-9576
• DOI: 10.1016/j.nima.2022.167779

The purpose of this study aimed to evaluate the noise reduction efficiency of a 3D median modified Wiener filter (MMWF) in brain T1-weighted magnetic resonance (MR) images. A simulation using BrainWeb phantom data and real experimental research based on data of the Alzheimer’s Disease Neuroimaging Initiative (ADNI) were performed, and the 2D MMWF was modeled to prove the usefulness of the proposed 3D MMWF. Brain MR images were obtained according to the kernel size and noise level of 2D and 3D MMWF, and the coefficient of variation (COV) and edge preservation index (EPI) were used for the quantitative evaluation of the obtained images. According to the changes in COV with respect to the changes in filter size, simulated T1-weighted images with 3D MMWF had a 2.76 times higher denoising performance than 2D MMWF. Furthermore, the EPI of simulated T1 weighted images with 3D MMWF had a 1.17 times better performance than that of simulated T1-weighted images with 2D MMWF, particularly in noisy images. To confirm the performance of 3D MMWF with clinical T1-weighted images, we obtained a T1-weighted image from ADNI and applied 2D and 3D MMWF with respect to kernel size. According to COV changes with respect to kernel size in both filters, clinical T1-weighted images showed a 1.14 times improvement with 3D MMWF and had a similar tendency as simulated images. We compared 2D and 3D MMWF in terms of tissue preservation and denoising performance in T1-weighted images. Our results indicate that the proposed 3D MMWF has better denoising performance than 2D MMWF for Rician noise and preserved the edges of brain tissues.