Data-Consistent Non-Cartesian Deep Subspace Learning for Efficient Dynamic MR Image Reconstruction

Self Cite

PurposeWidely used conventional 2D T2* approaches that are based on breath‐held, electrocardiogram (ECG)–gated, multi‐gradient‐echo sequences are prone to motion artifacts in the presence of incomplete breath holding or arrhythmias, which is common in cardiac patients. To address these limitations, a 3D, non‐ECG‐gated, free‐breathing T2* technique that enables rapid whole‐heart coverage was developed and validated.MethodsA continuous random Gaussian 3D k‐space sampling was implemented using a low‐rank tensor framework for motion‐resolved 3D T2* imaging. This approach was tested in healthy human volunteers and in swine before and after intravenous administration of ferumoxytol.ResultsSpatial‐resolution matched T2* images were acquired with 2–3‐fold reduction in scan time using the proposed T2* mapping approach relative to conventional T2* mapping. Compared with the conventional approach, T2* images acquired with the proposed method demonstrated reduced off‐resonance and flow artifacts, leading to higher image quality and lower coefficient of variation in T2*‐weighted images of the myocardium of swine and humans. Mean myocardial T2* values determined using the proposed and conventional approaches were highly correlated and showed minimal bias.ConclusionThe proposed non‐ECG‐gated, free‐breathing, 3D T2* imaging approach can be performed within 5 min or less. It can overcome critical image artifacts from undesirable cardiac and respiratory motion and bulk off‐resonance shifts at the heart–lung interface. The proposed approach is expected to facilitate faster and improved cardiac T2* mapping in those with limited breath‐holding capacity or arrhythmias.

Section: Discussionmentioning

confidence: 99%

Non‐electrocardiogram‐gated, free‐breathing, off‐resonance reduced, high‐resolution, whole‐heart myocardial T₂* mapping at 3 T within 5 min

Guan,

Yang,

Zhang

et al. 2024

Self Cite

“…[38][39][40][41][42][43] Deep-learning-based methods have also been applied to subspace reconstruction problems. [44][45][46][47] For example, a deep subspace learning method combined a deep network with a low-rank subspace modeling, thus allowing improved performance in T 1 mapping using a single-shot IR radial FLASH sequence. 46 However, those methods based on a supervised training paradigm demand fully sampled or reference data for training deep networks, which hinders their application in qMRI reconstruction problems where it may be prohibitively difficult to obtain fully sampled or reference data.…”

Section: Introductionmentioning

confidence: 99%

“…In MR image reconstruction, a combination of a deep‐learning‐based regularizer and parallel imaging forward model outperformed conventional parallel imaging and compressed sensing 38–43 . Deep‐learning‐based methods have also been applied to subspace reconstruction problems 44–47 . For example, a deep subspace learning method combined a deep network with a low‐rank subspace modeling, thus allowing improved performance in T 1 mapping using a single‐shot IR radial FLASH sequence 46 .…”

Section: Introductionmentioning

confidence: 99%

Zero‐DeepSub: Zero‐shot deep subspace reconstruction for rapid multiparametric quantitative MRI using 3D‐QALAS

Jun,

Arefeen,

Cho

et al. 2024

PurposeTo develop and evaluate methods for (1) reconstructing 3D‐quantification using an interleaved Look‐Locker acquisition sequence with T2 preparation pulse (3D‐QALAS) time‐series images using a low‐rank subspace method, which enables accurate and rapid T1 and T2 mapping, and (2) improving the fidelity of subspace QALAS by combining scan‐specific deep‐learning‐based reconstruction and subspace modeling.Theory and MethodsA low‐rank subspace method for 3D‐QALAS (i.e., subspace QALAS) and zero‐shot deep‐learning subspace method (i.e., Zero‐DeepSub) were proposed for rapid and high fidelity T1 and T2 mapping and time‐resolved imaging using 3D‐QALAS. Using an ISMRM/NIST system phantom, the accuracy and reproducibility of the T1 and T2 maps estimated using the proposed methods were evaluated by comparing them with reference techniques. The reconstruction performance of the proposed subspace QALAS using Zero‐DeepSub was evaluated in vivo and compared with conventional QALAS at high reduction factors of up to nine‐fold.ResultsPhantom experiments showed that subspace QALAS had good linearity with respect to the reference methods while reducing biases and improving precision compared to conventional QALAS, especially for T2 maps. Moreover, in vivo results demonstrated that subspace QALAS had better g‐factor maps and could reduce voxel blurring, noise, and artifacts compared to conventional QALAS and showed robust performance at up to nine‐fold acceleration with Zero‐DeepSub, which enabled whole‐brain T1, T2, and PD mapping at 1 mm isotropic resolution within 2 min of scan time.ConclusionThe proposed subspace QALAS along with Zero‐DeepSub enabled high fidelity and rapid whole‐brain multiparametric quantification and time‐resolved imaging.

“…11 SR models trained only on k-space-truncated labels do not account for this additional T 2 blurring, complicating prospective application to TSE images. Recently, physics-driven deep learning methods have shown their value for various tasks in MRI, [12][13][14][15][16][17] and could be an effective approach to deal with TSE T 2 blurring in MRI SR. Currently, physics-driven neural networks have shown promise for PSF correction, but they have not been used in conjunction with SR. 18 Another barrier to prospective SR in MRI has been the "magnitude degradation" (MD) process (k-space truncation of magnitude images) used in early SR network training efforts.…”

Section: Introductionmentioning

confidence: 99%

Physics‐informed deep learning for T2‐deblurred superresolution turbo spin echo MRI

Chen

Stapleton

et al. 2023

PurposeDeep learning superresolution (SR) is a promising approach to reduce MRI scan time without requiring custom sequences or iterative reconstruction. Previous deep learning SR approaches have generated low‐resolution training images by simple k‐space truncation, but this does not properly model in‐plane turbo spin echo (TSE) MRI resolution degradation, which has variable T2 relaxation effects in different k‐space regions. To fill this gap, we developed a T2‐deblurred deep learning SR method for the SR of 3D‐TSE images.MethodsA SR generative adversarial network was trained using physically realistic resolution degradation (asymmetric T2 weighting of raw high‐resolution k‐space data). For comparison, we trained the same network structure on previous degradation models without TSE physics modeling. We tested all models for both retrospective and prospective SR with 3 × 3 acceleration factor (in the two phase‐encoding directions) of genetically engineered mouse embryo model TSE‐MR images.ResultsThe proposed method can produce high‐quality 3 × 3 SR images for a typical 500‐slice volume with 6–7 mouse embryos. Because 3 × 3 SR was performed, the image acquisition time can be reduced from 15 h to 1.7 h. Compared to previous SR methods without TSE modeling, the proposed method achieved the best quantitative imaging metrics for both retrospective and prospective evaluations and achieved the best imaging‐quality expert scores for prospective evaluation.ConclusionThe proposed T2‐deblurring method improved accuracy and image quality of deep learning–based SR of TSE MRI. This method has the potential to accelerate TSE image acquisition by a factor of up to 9.