Adaptive denoising of 3D volumetric MR images using local variance based estimator. (May 2020)

• Record Type: Journal Article
• Authors: Das, Pabitra; Pal, Chandrajit; Chakrabarti, Amlan; Acharyya, Amit; Basu, Saumyajit
• Is Part Of: Biomedical signal processing and control. Volume 59(2020)
• Publication Date: 2020-05
• Subjects: Magnetic resonance imaging (MRI) -- Rician noise -- Denoising -- Noise estimation -- Adaptive diffusion-based MRI filtering (ADMF3D); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.101901
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The diffusion coefficient, controlling the averaging process has been made adaptive. A statistical edge stopping function is applied to preserve the detail of MRI images. A variance-based local noise estimator is used, influencing the diffusion process. Our filtering technique preserves the anatomical details of the cranial nerve. Abstract: Preservation of the anatomical structures during denoising of medical images is a very significant and challenging operation. Corruption of magnetic resonance image (MRI) by Rician noise is inherent to the acquisition process, affecting diagnosis. In this study, we present a novel filtering methodology that removes Rician noise from MRI by estimating the local noise variance, which drives the diffusion process of the filter. In our methodology, the adaptation of statistical edge stopping function captivates the preservation condition of the anatomical structure of the MRI images. The results obtained on synthetic/simulated MRI datasets (3D) and real MRI datasets confirm the accuracy and robustness of the proposed methodology. Compared to the benchmark approaches like BM4D, LTA3D, RNOLMMSE, ROLMMSE, MNL-tSVD and PRINLM3D, the optimized way of choosing the edge stopping functions, the automatic adjustment of the filtering coefficients and variance based local noise estimation technique lead to a qualitative and quantitative robust estimation performance, in case of both simulated and real datasets.