A regularized reconstruction pipeline for high‐definition diffusion MRI in challenging regions incorporating a per‐shot image correction

Abstract

Purpose Diffusion MRI is of interest for clinical research and diagnosis. Whereas high‐ resolution DWI/DTI is hard to achieve by single‐shot methods, interleaved acquisitions can deliver these if motion and/or folding artefacts are overcome. Thanks to its ability to provide zoomed, folding‐free images, spatially encoded MRI can fulfill these requirements. This is here coupled with a regularized reconstruction and parallel receive methods, to deliver a robust scheme for human DWI/DTI at mm and sub‐mm resolutions. Methods Each shot along the spatially encoded dimension was reconstructed separately to retrieve per‐shot phase maps. These shots, together with coil sensitivities, were combined with spatially encoded quadratic phase‐encoding matrices associated to each shot, into single global operators. Their originating images were then iteratively computed aided by l1 and l2 regularization methods. When needed, motion‐corrupted shots were discarded and replaced by redundant information arising from parallel imaging. Results Full‐brain DTI experiments at 1 mm and restricted brain DTIs with 0.75 mm nominal in‐plane resolutions were acquired and reconstructed successfully by the new scheme. These 3 Tesla spetiotemporally encoded results compared favorably with EPI counterparts based on segmented and selective excitation schemes provided with the scanner. Conclusion A new procedure for achieving high‐definition diffusion‐based MRI was developed and demonstrated.

Diffusion‐weighted breast MRI of malignancies with submillimeter resolution and immunity to artifacts by spatiotemporal encoding at 3T

“…To reconstruct the multiband SPEN scans a coil calibration was done with dedicated prescans, acquiring b ‐zero images for each breast separately. Based on these, coil sensitivity maps were calculated using ESPIRiT with the BART code package, and a conjugate gradient sensitivity‐encoding (CG‐SENSE) reconstruction was applied to disentangle the multiband information arising from both breasts. Motion artifacts that might arise were minimized by comparing smoothed phase maps from each scan’s image, and eliminating their differences via suitable phase corrections.…”

“…Motion artifacts that might arise were minimized by comparing smoothed phase maps from each scan’s image, and eliminating their differences via suitable phase corrections. A similar procedure was applied in the interleaved high‐resolution SPEN scans . From data processed in this fashion for different b ‐weighting values, ADC maps were obtained after suitably correcting for the pixel‐dependent influences arising from both diffusion and imaging gradients .…”

High‐resolution diffusion MRI studies of development in pregnant mice visualized by novel spatiotemporal encoding schemes

“…This difference enables one to overcome the SPEN artifacts by performing a referenceless identification and removal of the imperfections arising between even/odd data sets or between different interleaved shots. 34,35,39 Indeed, the phase shifts characterizing the low-resolution but fully unfolded images ρ j (x,y) associated with each even/odd set and/or each interleaved scan can be computed in a point-wise fashion by using one of the images as reference:…”

“…Slice thicknesses and readout bandwidths were in all cases 1 mm and 625 kHz; diffusion parameters were δ = 3.2 ms, Δ = 10 ms, diffusion gradient = 35 G/cm (b-weighting ≈ 850 s/ mm 2 ), three orthogonal G d orientations. Single-shot EPI parameters: T R /T E = 2000/26 ms, field of view (FOV) =30 × 30 mm 2 , data matrix = 96 × 96, extra reference scan for correcting even/odd correction and reversed PE polarity scans to overcome field inhomogeneities,[37][38][39] two averages. Singleshot SPEN: T R /T E = 2000/38 ms, FOV = 32 × 30 mm 2 , data matrix = 96 × 96, no extra scans, four averages.…”

Dynamic T2 mapping by multi‐spin‐echo spatiotemporal encoding

“…Figures and show the full pulse sequence and data processing flow chart used in this study. The sequence involved up to 4 nested loops (Figure ); from inner‐ to outer‐most these include an oscillating readout acquisition delivering the 2D SPEN image, an MSE loop encoding the T 2 information, a multislice loop extending this information into a third dimension, and an optional multishot loop used for eventual interleaving and thereby resolution improvement along the SPEN dimension . The corresponding data processing flow chart (Figure ) started with a Fourier transform along the readout dimension.…”