A robust adaptive sampling method for faster acquisition of MR images

Highly accelerated dynamic acquisition of 3D grid-tagged hyperpolarized-gas lung images using compressed sensing

PURPOSE

To develop and test compressed sensing-based multiframe 3D MRI of grid-tagged hyperpolarized gas in the lung.

THEORY AND METHODS

Applying grid-tagging RF pulses to inhaled hyperpolarized gas results in images in which signal intensity is predictably and sparsely distributed. In the present work, this phenomenon was used to produce a sampling pattern in which k-space is undersampled by a factor of approximately seven, yet regions of high k-space energy remain densely sampled. Three healthy subjects received multiframe 3D 3 He tagging MRI using this undersampling method. Images were collected during a single exhalation at eight timepoints spanning the breathing cycle from end-of-inhalation to end-of-exhalation. Grid-tagged images were used to generate 3D displacement maps of the lung during exhalation, and time-resolved maps of principal strains and fractional volume change were generated from these displacement maps using finite-element analysis.

RESULTS

Tags remained clearly resolvable for 4-6 timepoints (5-8 s) in each subject. Displacement maps revealed noteworthy temporal and spatial nonlinearities in lung motion during exhalation. Compressive normal strains occurred along all three principal directions but were primarily oriented in the head-foot direction. Fractional volume changes displayed clear bilateral symmetry, but with the lower lobes displaying slightly higher change than the upper lobes in 2 of the 3 subjects.

CONCLUSION

We developed a compressed sensing-based method for multiframe 3D MRI of grid-tagged hyperpolarized gas in the lung during exhalation. This method successfully overcomes previous challenges for 3D dynamic grid-tagging, allowing time-resolved biomechanical readouts of lung function to be generated.

Data-driven optimization of sampling patterns for MR brain T1ρ mapping

PURPOSE

The goal of this study was to apply a fast data-driven optimization algorithm, called bias-accelerated subset selection, for MR brain T1ρ mapping to generate optimized sampling patterns (SPs) for compressed sensing reconstruction of brain 3D-T1ρ MRI.

METHODS

Five healthy volunteers were recruited, and fully sampled Cartesian 3D-T1ρ MRIs were obtained. Variable density (VD) and Poisson disc (PD) undersampling was used as the input to SP optimization process. The reconstruction used 3 compressed sensing methods: spatiotemporal finite differences, low-rank plus sparse with spatial finite differences, and low rank. The performance of images and T1ρ maps using PD-SP and VD-SP and their optimized sampling patterns (PD-OSP and VD-OSP) were compared to the fully sampled reference using normalized root mean square error (NRMSE).

RESULTS

The VD-OSP with spatiotemporal finite differences reconstruction (NRMSE = 0.078) and the PD-OSP with spatiotemporal finite differences reconstruction (NRMSE = 0.079) at the highest acceleration factors (AF = 30) showed the largest improvement compared to the respective nonoptimized SPs (VD NRMSE = 0.087 and PD NRMSE = 0.149). Prospective undersampling was tested at AF = 4, with VD-OSP NRMSE = 0.057 versus PD-OSP NRMSE = 0.060, with optimized sampling performing better that input PD or VD sampling. For brain T1ρ mapping, the VD-OSP with low rank reconstruction for AFs <10 and VD-OSP with spatiotemporal finite differences for AFs >10 perform better.

CONCLUSIONS

The study demonstrated that the appropriate use of data-driven optimized sampling and suitable compressed sensing reconstruction technique can be employed to potentially accelerate 3D T1ρ mapping for brain imaging applications.

Alternating Learning Approach for Variational Networks and Undersampling Pattern in Parallel MRI Applications

This work proposes an alternating learning approach to learn the sampling pattern (SP) and the parameters of variational networks (VN) in accelerated parallel magnetic resonance imaging (MRI). We investigate four variations of the learning approach, that alternates between improving the SP, using bias-accelerated subset selection, and improving parameters of the VN, using ADAM. The variations include the use of monotone or non-monotone alternating steps and systematic reduction of learning rates. The algorithms learn an effective pair to be used in future scans, including an SP that captures fewer k-space samples in which the generated undersampling artifacts are removed by the VN reconstruction. The quality of the VNs and SPs obtained by the proposed approaches is compared against different methods, including other kinds of joint learning methods and state-of-art reconstructions, on two different datasets at various acceleration factors (AF). We observed improvements visually and in three different figures of merit commonly used in deep learning (RMSE, SSIM, and HFEN) on AFs from 2 to 20 with brain and knee joint datasets when compared to the other approaches. The improvements ranged from 1% to 62% over the next best approach tested with VNs. The proposed approach has shown stable performance, obtaining similar learned SPs under different initial training conditions. We observe that the improvement is not only due to the learned sampling density, it is also due to the learned position of samples in k-space. The proposed approach was able to learn effective pairs of SPs and reconstruction VNs, improving 3D Cartesian accelerated parallel MRI applications.

Fast data-driven learning of parallel MRI sampling patterns for large scale problems

In this study, a fast data-driven optimization approach, named bias-accelerated subset selection (BASS), is proposed for learning efficacious sampling patterns (SPs) with the purpose of reducing scan time in large-dimensional parallel MRI. BASS is applicable when Cartesian fully-sampled k-space measurements of specific anatomy are available for training and the reconstruction method for undersampled measurements is specified; such information is used to define the efficacy of any SP for recovering the values at the non-sampled k-space points. BASS produces a sequence of SPs with the aim of finding one of a specified size with (near) optimal efficacy. BASS was tested with five reconstruction methods for parallel MRI based on low-rankness and sparsity that allow a free choice of the SP. Three datasets were used for testing, two of high-resolution brain images ([Formula: see text]-weighted images and, respectively, [Formula: see text]-weighted images) and another of knee images for quantitative mapping of the cartilage. The proposed approach has low computational cost and fast convergence; in the tested cases it obtained SPs up to 50 times faster than the currently best greedy approach. Reconstruction quality increased by up to 45% over that provided by variable density and Poisson disk SPs, for the same scan time. Optionally, the scan time can be nearly halved without loss of reconstruction quality. Quantitative MRI and prospective accelerated MRI results show improvements. Compared with greedy approaches, BASS rapidly learns effective SPs for various reconstruction methods, using larger SPs and larger datasets; enabling better selection of sampling-reconstruction pairs for specific MRI problems.

Identification of sampling patterns for high-resolution compressed sensing MRI of porous materials: 'learning' from X-ray microcomputed tomography data

There exists a strong motivation to increase the spatial resolution of magnetic resonance imaging (MRI) acquisitions so that MRI can be used as a microscopy technique in the study of porous materials. This work introduces a method for identifying novel data sampling patterns to achieve undersampling schemes for compressed sensing MRI (CS-MRI) acquisitions, enabling 3D spatial resolutions of 17.6 µm to be achieved. A data-driven learning approach is used to derive k-space undersampling schemes for 3D MRI acquisitions from 3D X-ray microcomputed tomography (µCT) datasets acquired at a higher spatial resolution than can be acquired using MRI. The performance of the new sampling approach was compared to other, well-established sampling strategies using simulated MRI data obtained from high-resolution µCT images of rock core plugs. These simulations were performed for a range of different k-space sampling fractions (0.125-0.375) using images of Ketton limestone. The method was then extended to consideration of imaging Estaillades limestone and Fontainebleau sandstone. The results show that the new sampling approach performs as well as or better than conventional variable density sampling and without need for time-consuming parameter optimisation. Further, a bespoke sampling pattern is produced for each rock type. The novel undersampling strategy was employed to acquire 3D magnetic resonance images of a Ketton limestone rock at spatial resolutions of 35 and 17.6 µm. The ability of the k-space sampling scheme produced using the new approach in enabling reconstruction of the pore space characteristics of the rock was then demonstrated by benchmarking against the pore space statistics obtained from high-resolution µCT data. The MRI data acquired at 17.6 µm resolution gave excellent agreement with the pore size distribution obtained from the X-ray microcomputed tomography dataset, while the pore coordination number distribution obtained from the MRI data was slightly skewed to lower coordination numbers. This approach provides a method of producing a k-space undersampling pattern for MRI acquisition at a spatial resolution for which a fully sampled acquisition at that spatial resolution would be impractically long. The approach can be easily extended to other CS-MRI techniques, such as spatially resolved flow and relaxation time mapping. LAY DESCRIPTION: Magnetic resonance imaging (MRI) is widely used to study the microstructure of, and fluid transport phenomena in porous media relevant for engineering applications. A major application is the study of water and hydrocarbon transport in porous sedimentary rocks, which typically have pore sizes smaller than 100 µm. The spatial resolution of routine MRI acquisitions, however, is limited to several hundred µm due to the relatively low sensitivity of the magnetic resonance method. Therefore, there exists a strong motivation to increase the spatial resolution of MRI by one to two orders of magnitude to be able to study these rocks at a pore scale. This work reports the initial step towards achieving this. Three-dimensional images of rock pore structure are acquired at both 35 and 17.6 µm spatial resolution. In ongoing work, these methods are now being incorporated into magnetic resonance velocity imaging methods, thereby enabling imaging of both pore structure and hydrodynamics at these much higher spatial resolutions than were hitherto possible. Although X-ray microcomputed tomography (µCT) produces high spatial resolution images, it is far more limited in being able to spatially map transport processes (i.e. flow) in porous media. This work reports a strategy for accelerating the image acquisition time such that sufficient signal-to-noise ratio (SNR) is achieved to increase the spatial resolution, that is, the voxel size within which there is sufficient SNR within the resulting image. To achieve this, a technique known as compressed sensing is used which exploits undersampling of the acquired data relative to the standard fully sampled image. In MRI, data are acquired in so-called k-space and Fourier transformed to yield the real space image. The challenge, when undersampling, is to optimise the specific points in k-space that are acquired because these will influence the quality of the resulting image. This work reports a straightforward, robust strategy for identifying the optimal sets of k-space points to acquire. The method introduced uses simulated MRI images calculated from high-resolution µCT images of the rocks of interest, from which optimised MRI sampling patterns are obtained. The method does not require any optimisation of parameters for its implementation, which is a significant advantage compared to other strategies. Moreover, we show that the pore space characteristics of the acquired MRI images are in excellent agreement with the same characteristics obtained from a high-resolution µCT image.

Learning-Based Compressive MRI

In the area of magnetic resonance imaging (MRI), an extensive range of non-linear reconstruction algorithms has been proposed which can be used with general Fourier subsampling patterns. However, the design of these subsampling patterns has typically been considered in isolation from the reconstruction rule and the anatomy under consideration. In this paper, we propose a learning-based framework for optimizing MRI subsampling patterns for a specific reconstruction rule and anatomy, considering both the noiseless and noisy settings. Our learning algorithm has access to a representative set of training signals, and searches for a sampling pattern that performs well on average for the signals in this set. We present a novel parameter-free greedy mask selection method and show it to be effective for a variety of reconstruction rules and performance metrics. Moreover, we also support our numerical findings by providing a rigorous justification of our framework via statistical learning theory.