Rapid Dixon acquisitions for water/lipid separation in MRI.

摘要

The main limitation of current lipid/water suppression techniques in MRI is that these methods significantly increase the overall acquisition time of the particular sequence. For example, the multi-point Dixon methods utilize multiple acquisitions at different echo times to algebraically calculate separate fat and water images. Extended acquisition times result in an undesirable increase in respiratory and cardiac motion artifacts. Rapid acquisitions also reduce the duration of and potential errors in interventional MRI procedures thereby reducing the overall risk to the patient. In this work, rapid acquisition and image reconstruction techniques were developed to improve the temporal resolution of the conventional 2-Point Dixon (2PD) method for lipid/water separation.; A Keyhole Dixon acquisition was developed by combining a full K-space acquisition with a partial (i.e., keyhole) K-space acquisition. The number of acquired views for the centrally-symmetric keyhole acquisitions was optimized with a perceptual difference model (PDM) to sufficiently oversample the central region of K-space. The Keyhole Dixon technique resulted in a 25--38% reduction in the overall acquisition time relative to the 2PD acquisition for phantom and volunteer imaging studies with perceptual change in image quality.; Radial and rectilinear 1-Point Dixon (1PD) acquisitions were developed by applying the Dixon echo time variation between even and odd K-space lines. The oversampling of the central region of K-space inherent in radial acquisitions produced fat and water images from a single acquisition. For the rectilinear 1PD acquisition, a SENSE-like parallel imaging technique was used to separate the on-resonance water signal from the off-resonance lipid signal. Both 1PD acquisitions resulted in a 50% reduction in the 2PD acquisition time with comparable spatial resolution.; A genetic algorithm (GA) was used to create time-optimal 2-Point Dixon pulse sequences. The GA produced a Pareto-optimal series of pulse sequences at varying field-of-view (FOV) and readout bandwidth with the combined constraints of both gradient hardware and a vendor-specific peripheral nerve stimulation (PNS) model. The genetic optimization resulted in a 10--15% reduction in acquisition time in comparison to a standard dual-echo pulse sequence and a ∼50% reduction in comparison to a single-echo 2PD acquisition.