Faculty Bibliography

PURPOSE/OBJECTIVE:To develop a model-guided self-supervised deep learning MRI reconstruction framework called reference-free latent map extraction (RELAX) for rapid quantitative MR parameter mapping. METHODS:mapping in both simulated and actually acquired MRI data sets and was compared with supervised learning and conventional constrained reconstruction for suppressing noise and/or undersampling-induced artifacts. RESULTS:mapping data sets, RELAX achieved superior reconstruction quality compared with conventional iterative reconstruction, and similar reconstruction performance to supervised deep learning reconstruction. CONCLUSION/CONCLUSIONS:This work has demonstrated the initial feasibility of rapid quantitative MR parameter mapping based on self-supervised deep learning. The RELAX framework may also be further extended to other quantitative MRI applications by incorporating corresponding quantitative imaging models.

PURPOSE:Golden-angle single-shot PROPLLER (GA-SS-PROP) is proposed to accelerate the PROPELLER acquisition for distortion-free diffusion-weighted (DW) imaging. Acceleration is achieved by acquiring one-shot per b-value and several b-values can be acquired along a diffusion direction, where the DW signal follows a bi-exponential decay (i.e. IVIM). Sparse reconstruction is used to reconstruct full resolution DW images. Consequently, apparent diffusion coefficient (ADC) map and IVIM maps (i.e., perfusion fraction (f) and the perfusion-free diffusion coefficient (D)) are obtained simultaneously. The performance of GA-SS-PROP was demonstrated with simulation and human experiments. METHODS:A realistic numerical phantom of high-quality diffusion images of the brain was developed. The error of the reconstructed DW images and quantitative maps were compared to the ground truth. The pulse sequence was developed to acquire human brain data. For comparison, fully sampled PROPELLER and conventional single-shot echo planar imaging (SS-EPI) acquisitions were performed. RESULTS:). CONCLUSION:GA-SS-PROP offers fast, high-resolution and distortion-free DW images. The generated quantitative maps (f, D and ADC) can provide valuable information on tissue perfusion and diffusion properties simultaneously, which are desirable in many applications, especially in oncology. As a turbo spin-echo based technique, it can be applied in most challenging regions where SS-EPI is problematic.

Preclinical 4D flow MRI remains challenging and is restricted for parallel imaging acceleration due to the limited number of available receive channels. A radial acquisition with combined parallel imaging and temporal compressed sensing reconstruction was implemented to achieve accelerated preclinical 4D flow MRI. In order to increase the accuracy of the measured velocities, a quantitative evaluation of different temporal regularization weights for the compressed sensing reconstruction based on velocity instead of magnitude data is performed. A 3D radial retrospectively triggered phase contrast sequence with a combined parallel imaging and compressed sensing reconstruction with temporal regularization was developed. It was validated in a phantom and in vivo (C57BL/6 J mice), against an established fully sampled Cartesian sequence. Different undersampling factors (USFs [12, 15, 20, 30, 60]) were evaluated, and the effect of undersampling was analyzed in detail for magnitude and velocity data. Temporal regularization weights λ were evaluated for different USFs. Acceleration factors of up to 20 compared with full Nyquist sampling were achieved. The peak flow differences compared with the Cartesian measurement were the following: USF 12, 3.38%; USF 15, 4.68%; USF 20, 0.95%. The combination of 3D radial center-out trajectories and compressed sensing reconstruction is robust against motion and flow artifacts and can significantly reduce measurement time to 30 min at a resolution of 180 μm³ . Concisely, radial acquisition with combined compressed sensing and parallel imaging proved to be an excellent method for analyzing complex flow patterns in mice.

PURPOSE/OBJECTIVE:To develop and evaluate a deep adversarial learning-based image reconstruction approach for rapid and efficient MR parameter mapping. METHODS:mapping of the brain and the knee at an acceleration rate R = 8 and was compared with other state-of-the-art reconstruction methods. Global and regional quantitative assessments were performed to demonstrate the reconstruction performance of the proposed method. RESULTS:estimation. The quantitative metrics were normalized root mean square error of 3.6% for brain and 7.3% for knee, structural similarity index of 85.1% for brain and 83.2% for knee, and tenengrad measures of 9.2% for brain and 10.1% for the knee. The adversarial approach also achieved better performance for maintaining greater image texture and sharpness in comparison to the CNN approach without adversarial learning. CONCLUSION/CONCLUSIONS:The proposed framework by incorporating the efficient end-to-end CNN mapping, adversarial learning, and physical model enforced data consistency is a promising approach for rapid and efficient reconstruction of quantitative MR parameters.

PURPOSE:To propose a real-time 3D MRI technique called MR SIGnature MAtching (MRSIGMA) for high-resolution volumetric imaging and motion tracking with very low imaging latency. METHODS:MRSIGMA consists of two steps: (1) offline learning of a database of possible 3D motion states and corresponding motion signature ranges and (2) online matching of new motion signatures acquired in real time with prelearned motion states. Specifically, the offline learning step (non-real-time) reconstructs motion-resolved 4D images representing different motion states and assigns a unique motion range to each state. The online matching step (real-time) acquires motion signatures only and selects one of the prelearned 3D motion states for each newly acquired signature, which generates 3D images efficiently in real time. The MRSIGMA technique was evaluated on 15 golden-angle stack-of-stars liver data sets, and the performance of respiratory motion tracking with the online-generated real-time 3D MRI was compared with the corresponding 2D projections acquired in real time. RESULTS: = 0.948) between motion displacement measured from the online-generated real-time 3D images and the 2D real-time projections. CONCLUSION:This proof-of-concept study demonstrates the feasibility of MRSIGMA for high-resolution real-time volumetric imaging, which shifts the acquisition and reconstruction burden to an offline learning step and leaves fast online matching for online imaging with very low imaging latency. The MRSIGMA technique can potentially be used for real-time motion tracking in MRI-guided radiation therapy.

PURPOSE:The purpose of this study was to evaluate the performance of motion-weighted Golden-angle RAdial Sparse Parallel MRI (motion-weighted GRASP) for free-breathing dynamic contrast-enhanced MRI (DCE-MRI) of the lung. METHODS:-weighted stack-of-stars golden-angle radial sequence and a post-contrast breath-hold MR scan using a Cartesian volumetric-interpolated imaging sequence (BH-VIBE). Each radial dataset was reconstructed using GRASP without motion compensation and motion-weighted GRASP. All MR images were visually evaluated by two experienced radiologists blinded to reconstruction and acquisition schemes independently. In addition, the influence of motion-weighted reconstruction on dynamic contrast-enhancement patterns was also investigated. RESULTS:For image quality assessment, motion-weighted GRASP received significantly higher visual scores than GRASP (P < 0.05) for overall image quality (3.68 vs. 3.39), lesion conspicuity (3.54 vs. 3.18) and overall artifact level (3.53 vs. 3.15). There was no significant difference (P > 0.05) between the breath-hold BH-VIBE and motion-weighted GRASP images. For assessment of temporal fidelity, motion-weighted GRASP maintained a good agreement with respect to GRASP. CONCLUSION:Motion-weighted GRASP achieved better reconstruction performance in free-breathing DCE-MRI of the lung compared to standard GRASP, and it may enable improved assessment of pulmonary lesions.

Retrospective electrocardiogram-gated, 2D phase-contrast (PC) flow MRI is routinely used in clinical evaluation of valvular/vascular disease in pediatric patients with congenital heart disease (CHD). In patients not requiring general anesthesia, clinical standard PC is conducted with free breathing for several minutes per slice with averaging. In younger patients under general anesthesia, clinical standard PC is conducted with breath-holding. One approach to overcome this limitation is using either navigator gating or self-navigation of respiratory motion, at the expense of lengthening scan times. An alternative approach is using highly accelerated, free-breathing, real-time PC (rt-PC) MRI, which to date has not been evaluated in CHD patients. The purpose of this study was to develop a 38.4-fold accelerated 2D rt-PC pulse sequence using radial k-space sampling and compressed sensing with 1.5 × 1.5 × 6.0 mm³ nominal spatial resolution and 40 ms nominal temporal resolution, and evaluate whether it is capable of accurately measuring flow in 17 pediatric patients (aortic valve, pulmonary valve, right and left pulmonary arteries) compared with clinical standard 2D PC (either breath-hold or free breathing). For clinical translation, we implemented an integrated reconstruction pipeline capable of producing DICOMs of the order of 2 min per time series (46 frames). In terms of association, forward volume, backward volume, regurgitant fraction, and peak velocity at peak systole measured with standard PC and rt-PC were strongly correlated (R² > 0.76; P < 0.001). Compared with clinical standard PC, in terms of agreement, forward volume (mean difference = 1.4% (3.0% of mean)) and regurgitant fraction (mean difference = -2.5%) were in good agreement, whereas backward volume (mean difference = -1.1 mL (28.2% of mean)) and peak-velocity at peak systole (mean difference = -21.3 cm/s (17.2% of mean)) were underestimated by rt-PC. This study demonstrates that the proposed rt-PC with the said spatial resolution and temporal resolution produces relatively accurate forward volumes and regurgitant fractions but underestimates backward volumes and peak velocities at peak systole in pediatric patients with CHD.

PURPOSE/OBJECTIVE:To propose a highly accelerated, high-resolution dynamic contrast-enhanced MRI (DCE-MRI) technique called GRASP-Pro (golden-angle radial sparse parallel imaging with imProved performance) through a joint sparsity and self-calibrating subspace constraint with automated selection of contrast phases. METHODS:GRASP-Pro reconstruction enforces a combination of an explicit low-rank subspace-constraint and a temporal sparsity constraint. The temporal basis used to construct the subspace is learned from an intermediate reconstruction step using the low-resolution portion of radial k-space, which eliminates the need for generating the basis using auxiliary data or a physical signal model. A convolutional neural network was trained to generate the contrast enhancement curve in the artery, from which clinically relevant contrast phases are automatically selected for evaluation. The performance of GRASP-Pro was demonstrated for high spatiotemporal resolution DCE-MRI of the prostate and was compared against standard GRASP in terms of overall image quality, image sharpness, and residual streaks and/or noise level. RESULTS:Compared to GRASP, GRASP-Pro reconstructed dynamic images with enhanced sharpness, less residual streaks and/or noise, and finer delineation of the prostate without prolonging reconstruction time. The image quality improvement reached statistical significance (P < 0.05) in all the assessment categories. The neural network successfully generated contrast enhancement curves in the artery, and corresponding peak enhancement indexes correlated well with that from the manual selection. CONCLUSION/CONCLUSIONS:GRASP-Pro is a promising method for rapid and continuous DCE-MRI. It enables superior reconstruction performance over standard GRASP and allows reliable generation of artery enhancement curve to guide the selection of desired contrast phases for improving the efficiency of GRASP MRI workflow.

BACKGROUND:Quantifying the biomechanical properties of pancreatic tumors could potentially help with assessment of tumor aggressiveness, prognosis, and prediction of therapy response. PURPOSE/OBJECTIVE:To quantify respiratory-induced deformation in the pancreas and pancreatic lesions using XD-GRASP (eXtra-Dimensional Golden-angle RAdial Sparse Parallel), MRI. STUDY TYPE/METHODS:W) imaging were studied. SUBJECTS/METHODS:Thirty-two patients (12 male and 20 female) including nine with pancreatic lesions constituted our study cohort. FIELD STRENGTH/SEQUENCE/UNASSIGNED:WI contrast-enhanced gradient echo radial free-breathing acquisition. ASSESSMENT/RESULTS:Using the XD-GRASP imaging technique, the acquired free-breathing radial data were sorted and binned into 10 consecutive respiratory motion states that were jointly reconstructed. 3D deformation fields along the respiratory dimension were computed using an optical flow method and were analyzed in the pancreas. STATISTICAL TESTS/UNASSIGNED:The Wilcoxon signed-rank test was used to assess the difference in average displacement across pancreatic regions, while the Wilcoxon rank-sum test was used for displacement differences between patients with and without tumors. The interclass correlation coefficient (ICC) was computed to assess consistency between observers for each image quality measure. RESULTS:There was a significantly larger displacement in the pancreatic tail compared with the head (8.2 ± 3.7 mm > 5.8 ± 2.4 mm; P < 0.001) and body regions (8.2 ± 3.7 mm > 6.6 ± 2.9 mm; P < 0.001). Furthermore, there was reduced normalized average displacement in patients with pancreatic lesions compared with subjects without lesions (0.33 ± 0.1 < 0.69 ± 0.26, P < 0.001 for the head; 0.30 ± 0.1 < 0.84 ± 0.31, P < 0.001 for the body; and 0.44 ± 0.31 < 1.08 ± 0.53, P < 0.001 for the tail, respectively). DATA CONCLUSION/UNASSIGNED:Free-breathing respiratory motion-sorted XD-GRASP MRI has the potential to noninvasively characterize the biomechanical properties of the pancreas by quantifying breathing-induced mechanical displacement. LEVEL OF EVIDENCE/METHODS:4 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2019.

PURPOSE:To develop and evaluate a novel deep learning-based reconstruction framework called SANTIS (Sampling-Augmented Neural neTwork with Incoherent Structure) for efficient MR image reconstruction with improved robustness against sampling pattern discrepancy. METHODS:With a combination of data cycle-consistent adversarial network, end-to-end convolutional neural network mapping, and data fidelity enforcement for reconstructing undersampled MR data, SANTIS additionally utilizes a sampling-augmented training strategy by extensively varying undersampling patterns during training, so that the network is capable of learning various aliasing structures and thereby removing undersampling artifacts more effectively and robustly. The performance of SANTIS was demonstrated for accelerated knee imaging and liver imaging using a Cartesian trajectory and a golden-angle radial trajectory, respectively. Quantitative metrics were used to assess its performance against different references. The feasibility of SANTIS in reconstructing dynamic contrast-enhanced images was also demonstrated using transfer learning. RESULTS:Compared to conventional reconstruction that exploits image sparsity, SANTIS achieved consistently improved reconstruction performance (lower errors and greater image sharpness). Compared to standard learning-based methods without sampling augmentation (e.g., training with a fixed undersampling pattern), SANTIS provides comparable reconstruction performance, but significantly improved robustness, against sampling pattern discrepancy. SANTIS also achieved encouraging results for reconstructing liver images acquired at different contrast phases. CONCLUSION:By extensively varying undersampling patterns, the sampling-augmented training strategy in SANTIS can remove undersampling artifacts more robustly. The novel concept behind SANTIS can particularly be useful for improving the robustness of deep learning-based image reconstruction against discrepancy between training and inference, an important, but currently less explored, topic.