Correlated noise in brain magnetic resonance elastography

Abstract

Magnetic resonance elastography (MRE) is a phasecontrast MRI technique for creating in vivo mechanical property maps of tissues such as the brain. [1][2][3] Applied mechanical vibrations cause small-scale tissue deformations that are used to estimate brain viscoelastic mechanical properties, such as the shear stiffness and damping ratio, which are used to indirectly assess brain health. [4][5][6] Changes in these mechanical properties have been associated with demyelination, 7 inflammation, 8 neuronal loss, 9 and have been detected in development and aging 10,11

OSCILLATE: A low‐rank approach for accelerated magnetic resonance elastography

“…We ultimately determined that considering MEG directions as part of the temporal dimension, such that the full rank of the dataset is L = 24, would be better suited for the MRE problem as it gives more versatility in allowing for undersampling patterns that could be distributed across the repetitions. This reduces the importance on any single repetition in the joint image reconstruction, which is beneficial in the event of unexpected subject motion or physiological noise 44 …”

“…This reduces the importance on any single repetition in the joint image reconstruction, which is beneficial in the event of unexpected subject motion or physiological noise. 44 A key aspect of a low-rank reconstruction is the accurate determination of temporal basis functions, which describe how the image repetitions are related. Navigator images are a popular MRI tool used in multishot sequences to correct for motion-induced phase errors.…”

“…These navigator images have the same phase and image contrast as the MRE image data to which they belong and, thus, also the same temporal basis. Nonlinear motion‐induced phase error correction is achieved via the phase corrected SENSE algorithm, which applies the conjugate of the navigator phase to each associated shot during image reconstruction via the use of a shot specific SENSE map incorporating the conjugate of the error measured by the navigator phase, 50 which reduces shot‐to‐shot variation in phase arising from spurious vibrations and physiological motions 44 . Data are reconstructed accounting for B 0 field inhomogeneities via a fieldmap collected immediately prior to MRE data collection 51,52 .…”

“…Sorted in this order, and with $$$ {V}^H $$$ = $$$ \Phi $$$, we can truncate the rank of C to L , thereby retaining only the bases contributing the most energy to the signal. The remaining energy and bases primarily consist of random noise and information with less well‐defined spatiotemporal coherence, such as would arise from spurious vibrations and physiological motions present in brain MRE data 44 and, thus, do not add additional useful information to the MRE displacement image reconstruction. Ultimately, by reconstructing information at a low‐dimensional temporal subspace, less k‐space sampling is necessary to recover the entire set of images.…”

“…Nonlinear motion-induced phase error correction is achieved via the phase corrected SENSE algorithm, which applies the conjugate of the navigator phase to each associated shot during image reconstruction via the use of a shot specific SENSE map incorporating the conjugate of the error measured by the navigator phase, 50 which reduces shot-to-shot variation in phase arising from spurious vibrations and physiological motions. 44 Data are reconstructed accounting for B 0 field inhomogeneities via a fieldmap collected immediately prior to MRE data collection. 51,52 Images are reconstructed iteratively using a preconditioned conjugate gradient solver implementing non-Cartesian SENSE 47 with field inhomogeneity correction via time segmentation with one time segment per 2.8 ms of readout length, using the non-uniform fast Fourier transform (NUFFT).…”