Model-based iterative reconstruction for direct imaging with point spread function encoded echo planar MRI

Abstract

Background: Echo planar imaging (EPI) is a fast measurement technique commonly used in magnetic resonance imaging (MRI), but is highly sensitive to measurement non-idealities in reconstruction. Point spread function (PSF)-encoded EPI is a multi-shot strategy which alleviates distortion, but acquisition of encodings suitable for direct distortion-free imaging prolongs scan time. In this work, a model-based iterative reconstruction (MBIR) framework is introduced for direct imaging with PSF-EPI to improve image quality and acceleration potential.

Methods: An MBIR platform was developed for accelerated PSF-EPI. The reconstruction utilizes a subspace representation, is regularized to promote local low-rankedness (LLR), and uses variable splitting for efficient iteration. Comparisons were made against standard reconstructions from prospectively accelerated PSF-EPI data and with retrospective subsampling. Exploring aggressive partial Fourier acceleration of the PSF-encoding dimension, additional comparisons were made against an extension of Homodyne to direct PSF-EPI in numerical experiments. A neuroradiologists' assessment was completed comparing images reconstructed with MBIR from retrospectively truncated data directly against images obtained with standard reconstructions from non-truncated datasets.

Results: Image quality results were consistently superior for MBIR relative to standard and Homodyne reconstructions. As the MBIR signal model and reconstruction allow for arbitrary sampling of the PSF space, random sampling of the PSF-encoding dimension was also demonstrated, with quantitative assessments indicating best performance achieved through nonuniform PSF sampling combined with partial Fourier. With retrospective subsampling, MBIR reconstructs high-quality images from sub-minute scan datasets. MBIR was shown to be superior in a neuroradiologists' assessment with respect to three of five performance criteria, with equivalence for the remaining two.

Conclusions: A novel image reconstruction framework is introduced for direct imaging with PSF-EPI, enabling arbitrary PSF space sampling and reconstruction of diagnostic-quality images from highly accelerated PSF-encoded EPI data.

Keywords: Echo planar imaging; Image reconstruction; Low-rank; Point spread function.

Figure 1:

Pulse sequence diagram and sampling. A, Abbreviated pulse sequence diagram for spin-echo EPI with PSF mapping. Pass-specific auxiliary gradient encoding precedes each readout. Slice selection and optional spoiler gradients are omitted for clarity. B, PSF-EPI sampling diagram for a single channel. Red dots represent acquired readouts; light blue dots, unacquired samples. Acceleration along the PSF axis is uniform with rFOV acceleration combined with partial Fourier along PSF. C, PSF-EPI sampling diagram with random encoding and partial Fourier along the PSF axis.

Figure 2:

Comparison of identically windowed reconstructions from prospectively accelerated PSF-EPI acquisition with 28 shots (7/8 partial Fourier, 2:09 m:s scan duration) with further subsampling to 18 shots (9/16, 1:23 m:s). Left column, ordinary reconstruction; middle column, PSF-HD; right column, MBIR. Difference maps are shown between ordinary and PSF-HD, and between PSF-HD and MBIR, indicating moderate recovery of sharpness for PSF-HD relative to ordinary. MBIR shows substantial suppression of intracranial artifacts and sharpness recovery relative to PSF-HD. Extracranial globe ghost artifacts (red arrows) are exhibited by all reconstructions; however, intracranial ghosts (blue arrows) are suppressed with MBIR only. Insets of the brainstem at 18 shots demonstrate improved visualization via MBIR of cerebellar folds (purple arrows) and the aqueduct of Sylvius (green) despite acceleration. Bottom row shows difference maps between 28 and 18 shots for each reconstruction, showing loss of resolution for ordinary, and preservation of sharpness but subtle ringing artifacts in the temporal lobes with reduced shots for PSF-HD (yellow arrows).

Figure 3:

Additional comparison of identically windowed reconstructions as shown in Figure 2. Again PSF-HD shows sharpness recovery relative to zero-filled reconstruction, but with residual artifacts and increased visual noise. MBIR shows reduced intracranial artifacts and sharpness recovery relative to PSF-HD and ordinary. Globe ghosts (red arrows) are prominent for PSF-HD and ordinary reconstructions, and reduced for MBIR. Brainstem insets at 18 shots demonstrate suppressed ghosting anterior to the basilar artery (green arrows) via MBIR. Cerebellar folds (purple arrows) and the cisternal segment of the trigeminal nerve and adjacent structure (blue) are better visualized with MBIR. Bottom row shows difference maps between 28 and 18 shots for each reconstruction, showing loss of resolution for ordinary at cerebellar folds.

Figure 4:

Reconstruction comparison following retrospective subsampling to 12 shots (0:56 m:s simulated scan time). Top row and at bottom left: rFOV = 12x with partial Fourier in PSF. At bottom right: PSF encodes randomly sampled in a 1D Poisson disk with partial Fourier. rFOV-based reconstructions show signal dropout and anterior ghost artifact respective to a region of susceptibility (red arrows) and extracranial ghosts posteriorly (blue), both reduced for MBIR. Artifacts are reduced for MBIR with random sampling.

Figure 5:

Performance characterization across 100 reconstructions for 4 random sampling patterns: 1D Poisson disk (PD) with/out partial Fourier (PF) along PSF, and uniform random (UR) with/out PF. Maps are identically windowed. PD and UR both benefit in mean map quality from PF, with streak artifact in both symmetric maps (red arrows). Variance is increased in both variants for symmetric encoding and decreased with PF; 1D-PD with PF achieves lowest overall variance.

Figure 6:

Unblinded score count histograms for all evaluation criteria. Individual colors show individual readers’ score counts.

Figure 7:

Image comparison example from neuroradiologists’ assessment. Control images were reconstructed from full 28-shot datasets; MBIR, from 18-shot datasets. Control shows ghosting artifacts suppressed by MBIR (red arrows). MBIR shows sharper high-contrast structural detail in the intraventricular choroid plexus and cerebellar folds (magenta arrows), with enhanced visualization of low-contrast structures adjacent to ventricles and in the brainstem (blue arrows).