Five‐dimensional quantitative low‐dose Multitasking dynamic contrast‐ enhanced MRI: Preliminary study on breast cancer

Abstract

To develop a low-dose Multitasking DCE technique (LD-MT-DCE) for breast imaging, enabling dynamic T 1 mapping-based quantitative characterization of tumor blood flow and vascular properties with whole-breast coverage, a spatial resolution of 0.9 × 0.9 × 1.1 mm 3 , and a temporal resolution of 1.4 seconds using a 20% gadolinium dose (0.02 mmol/kg). Methods: Magnetic resonance Multitasking was used to reconstruct 5D images with three spatial dimensions, one T 1 recovery dimension for dynamic T 1 quantification, and one DCE dimension for contrast kinetics. Kinetic parameters F p , v p , K trans , and v e were estimated from dynamic T 1 maps using the two-compartment exchange model. The LD-MT-DCE repeatability and agreement against standard-dose MT-DCE were evaluated in 20 healthy subjects. In 7 patients with triple-negative breast cancer, LD-MT-DCE image quality and diagnostic results were compared with that of standarddose clinical DCE in the same imaging session. One-way unbalanced analysis of variance with Tukey test was performed to evaluate the statistical significance of the kinetic parameters between control and patient groups. Results: The LD-MT-DCE technique was repeatable, agreed with standard-dose MT-DCE, and showed excellent image quality. The diagnosis using LD-MT-DCE matched well with clinical results. The values of F p , v p , and K trans were significantly different between malignant tumors and normal breast tissue (P < .001, < .001, and < .001, respectively), and between malignant and benign tumors (P = .020, .003, and < .001, respectively). Conclusion: The LD-MT-DCE technique was repeatable and showed excellent image quality and equivalent diagnosis compared with standard-dose clinical DCE.

Motion‐robust quantitative multiparametric brain MRI with motion‐resolved MR multitasking

“…20 Note that it is possible to incorporate the multi-contrast model into the motion clustering algorithm to handle the dynamically varying image contrasts, which allows motion states to be resolved at motion time scales shorter than TR (eg, in the heart). 18 It is worth mentioning that, although this work focused on motion-resolved T 1 /T 2 /T 1ρ mapping as an application, the proposed method could be extended to other tissue parameter combinations that are available with the multitasking framework, including but not limited to T 1 /T 2 / ADC, 21 T 1 /T 2 /T 1ρ /T * 2 /QSM, 36 T 1 /T * 2 /proton density fat fraction, 37 CEST, 38 and perfusion and vascular permeability parameters with DCE MRI, 20,23,24 with proper sequence modification but without changing the reconstruction pipeline. We aim to provide a unified framework generalizable enough to not only different head motion patterns but also different tissue parameters in various clinical applications.…”

“…The most common sequences are magnetization preparation‐based (ie, T 2 ‐preparation, T 1 ρ ‐preparation, diffusion‐preparation, etc.) saturation recovery or inversion recovery sequences 18‐24 . FLASH readouts fill the recovery period between 2 preparations.…”

“…saturation recovery or inversion recovery sequences. [18][19][20][21][22][23][24] FLASH readouts fill the recovery period between 2 preparations. K-space data are continuously collected, and the entire dataset of k-space lines can be seen as the union of 2 nonoverlapping subsets: 1) auxiliary training data D tr are periodically collected every a few readouts either at the center k-space line for Cartesian sampling, or using the 0° spoke for radial sampling for motion identification and the estimation of [18][19][20][21][22][23][24] ; 2) the rest of the k-space lines belongs to imaging data D img , which traverses the entire k-space with either Gaussian-density random Cartesian trajectory or stackof-stars trajectory for the estimation of U.…”

“…20 In organs with less motion, such as carotid arteries, breast, and brain, abrupt motion detection and motion rejection have also been efficiently incorporated into this. [21][22][23][24] No external motionmonitoring devices are required with this framework.…”

Whole‐brain steady‐state CEST at 3 T using MR Multitasking

“…where × i denotes the tensor i-mode product; U x ∈ ℂ J×L 0 contains L 0 spatial basis functions with J voxels each; U Δ ∈ ℂ K×L 1 contains L 1 basis functions, which characterize the Z-spectra; U ∈ ℂ M×L 2 contains L 2 temporal basis functions, which characterize the signal evolution to reach steady state within each frequency offset; and ∈ ℂ L 0 ×L 1 ×L 2 denotes the core tensor. The core tensor and temporal bases can be combined into a temporal factor tensor Φ = × 2 U Δ × 3 U , in which case Equation (1) simplifies to Image reconstruction was done similarly to previous MR Multitasking works, [28][29][30][31] in two steps. First, the components of the temporal factor tensor Φ were estimated from the training data d tr .…”