A novel method for removing Rician noise from MRI based on variational mode decomposition. (August 2021)

• Record Type: Journal Article
• Authors: Pankaj, Divya; D., Govind; K.A., Narayanankutty
• Is Part Of: Biomedical signal processing and control. Volume 69(2021)
• Publication Date: 2021-08
• Subjects: Magnetic resonance imaging (MRI) -- Denoising -- Rician noise -- 2D-Variational mode decomposition -- 2D-VMD-TV method; 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.2021.102737
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Algorithm proposed for removing Rician noise characteristics from MRI. Variational mode decomposition (VMD) captures all frequency variations in images. Proposed method decomposed noisy MRI into its frequency components called modes. Eliminated the higher modes and reconstructed the denoised image using lower modes. In second stage, remnant noise details are removed using TV regularization method. Abstract: MRI is a highly efficient medical imaging modality that captures all the delicate structural features of human tissues and organs enabling accurate clinical diagnosis. The noise occurred during image acquisition and transmission degrades these structural features and misleads the clinical diagnosis. Rician is the thermal noise found in MRI due to the thermal agitation of electrons. The main objective of the proposed work is to eliminate the Rician noise characteristics from the MRI. The proposed method effectively utilizes the properties of an efficient decomposition technique called variational mode decomposition (VMD). The proposed denoising method is performed in two stages. The high-frequency modes from the decomposed image are discarded and the denoised image is reconstructed from the low-frequency modes. The total variation (TV) image smoothing based on non-convex optimization is performed in the second stage for removing the remnant noise details from the first stage. The effectiveness of the proposed work is confirmed based on the improved peak signal to noise ratio (PSNR), stuctural similarity index measure (SSIM), quality index based on local variance (QILV) and Bhattacharrya coefficient (BC) scores. The proposed method is compared with different existing denoising methods. The experiment is conducted on two different datasets such as simulated Brainweb database and clinical dataset. Based on the comparative experimental analysis, the effectiveness of the proposed VMD-TV is confirmed by the improved objective performance metrics. In addition to objective quality assessments, the subjective evaluations carried out by radiologist and neurologists show the relatively better visual quality of the proposed method compared to the methods such as linear minimum mean square error (LMMSE) and bilateral filtering (BF).