Denoising four-dimensional flow magnetic resonance imaging data using a split-and-overlap approach via singular value decomposition

Abstract

Four-dimensional flow magnetic resonance imaging (4D flow MRI) offers a powerful tool for visualizing fluid flows, critical for both diagnosing cardiovascular diseases and analyzing engineering fluid dynamics. Despite its potential in medical research, the clinical applicability of 4D flow MRI often faces challenges due to inherent noise. To mitigate this, we introduce the split-and-overlap singular value decomposition (SOSVD) filter, a distinctive noise reduction approach. Unlike traditional singular value decomposition methods, the SOSVD filter partitions the primary data matrix into overlapping subdomains and then applies singular value decomposition to each subdomain, preserving only the dominant mode for noise attenuation. Evaluations on simulated and experimental flow data within a square duct revealed a significant decrease in root mean square noise metrics. Moreover, when applied to in vivo aortic data, the SOSVD filter enhanced various flow determinants, including divergence, velocity gradients, streamlines, and velocity coherence. Thus, the SOSVD method presents a promising avenue for augmenting noise reduction in 4D flow MRI, potentially elevating diagnostic accuracy and enriching cardiovascular disease research.