Blind Compressed Sensing Dynamic MRI
暂无分享,去创建一个
Introduction: Achieving high spatio-temporal resolutions in dynamic MRI (DMRI) (eg. myocardial perfusion MRI) is often challenging due to the slow nature of MR acquisitions. Recently, several schemes that exploit the low-rank property of dynamic datasets were introduced to accelerate dynamic MRI [eg: 1-3]. These methods exploit the similarity of the voxel time profiles (intensity variations as a function of time) by expressing them as a linear combination of a few orthogonal temporal basis functions. Since the temporal bases and their coefficients (spatial weights) are estimated from the under-sampled Fourier data itself, this representation is termed as the blind linear model (BLM). This method provides good image quality at high accelerations, when the inter-frame motion is not very significant. However, the similarity of voxel profiles often degrades significantly with inter-frame motion (eg. free breathing perfusion). Since more basis functions, and equivalently more coefficients, are required to accurately represent the resulting dataset using BLM, the maximum acceleration that can be achieved using BLM degrades significantly with inter-frame motion.