Rapid three-dimensional multiparametric MRI with quantitative transient-state imaging

Abstract

Novel methods for quantitative, transient-state multiparametric imaging are increasingly being demonstrated for assessment of disease and treatment efficacy. Here, we build on these by assessing the most common Non-Cartesian readout trajectories (2D/3D radials and spirals), demonstrating efficient anti-aliasing with a k- space view-sharing technique, and proposing novel methods for parameter inference with neural networks that incorporate the estimation of proton density. Our results show good agreement with gold standard and phantom references for all readout trajectories at 1.5 T and 3 T. Parameters inferred with the neural network were within 6.58% difference from the parameters inferred with a high-resolution dictionary. Concordance correlation coefficients were above 0.92 and the normalized root mean squared error ranged between 4.2 and 12.7% with respect to gold-standard phantom references for T1 and T2. In vivo acquisitions demonstrate sub-millimetric isotropic resolution in under five minutes with reconstruction and inference times < 7 min. Our 3D quantitative transient-state imaging approach could enable high-resolution multiparametric tissue quantification within clinically acceptable acquisition and reconstruction times.

Retrospective rigid motion correction of three‐dimensional magnetic resonance fingerprinting of the human brain

“…Our method took 18 minutes per subject using an Intel Xeon processor E5‐2600 v4 (48 CPU cores) equipped with a NVIDIA Tesla K80 GPU. All gridding and inference were run on the GPU as described in Gòmez et al 29 …”

“…We compared different k ‐space trajectory orderings (shown in Figure 1 C‐E) in terms of motion robustness. In all our trajectories, we considered the dataset as a combination of 56 consecutive segments each consisting of 880 pulses 29 . All segments were acquired with the same FA schedule (see Figure 1A), preceded by an adiabatic inversion pulse.…”

“…In all our trajectories, we considered the dataset as a combination of 56 consecutive segments each consisting of 880 pulses. 29 All segments were acquired with the same FA schedule (see Figure 1A), preceded by an adiabatic inversion pulse. The identical FA schedule per segment allowed to create 56 equal contrast navigators used for realigning the data.…”

Repeatability and reproducibility of human brain morphometry using three‐dimensional magnetic resonance fingerprinting

“…The repeatability reported in this study could potentially be improved using an optimized acquisition schedule, a more advanced reconstruction algorithm (Asslander et al, 2018; Bustin et al, 2019; Pierre, Ma, Chen, Badve, & Griswold, 2016) or by reducing the discretization of dictionary using a finer step size or dictionary‐free neural network based inference (Cohen, Zhu, & Rosen, 2018; Golbabaee, Chen, Gómez, Menzel, & Davies, 2019; Gómez et al, 2020; Virtue, Yu, & Lustig, 2017). Additionally, the use of multichannel inputs (e.g., T1 and T2 maps), which could be obtained from a single 3D MRF sequence scan, could potentially improve the robustness of segmentation algorithms that rely only on T1‐weighted images.…”

“…The subjects were repositioned between the scan and rescan tests. The MRF data acquisition and reconstruction was performed as described by Gómez et al (Gómez et al, 2020). Briefly, the 3D MRF sequence was based on steady‐state free precession (SSFP) with a 3D spiral projection trajectory.…”