Showing papers by "Daniel K. Sodickson published in 2019"

Assessment of the generalization of learned image reconstruction and the potential for transfer learning

Abstract: Purpose Although deep learning has shown great promise for MR image reconstruction, an open question regarding the success of this approach is the robustness in the case of deviations between training and test data. The goal of this study is to assess the influence of image contrast, SNR, and image content on the generalization of learned image reconstruction, and to demonstrate the potential for transfer learning. Methods Reconstructions were trained from undersampled data using data sets with varying SNR, sampling pattern, image contrast, and synthetic data generated from a public image database. The performance of the trained reconstructions was evaluated on 10 in vivo patient knee MRI acquisitions from 2 different pulse sequences that were not used during training. Transfer learning was evaluated by fine-tuning baseline trainings from synthetic data with a small subset of in vivo MR training data. Results Deviations in SNR between training and testing led to substantial decreases in reconstruction image quality, whereas image contrast was less relevant. Trainings from heterogeneous training data generalized well toward the test data with a range of acquisition parameters. Trainings from synthetic, non-MR image data showed residual aliasing artifacts, which could be removed by transfer learning-inspired fine-tuning. Conclusion This study presents insights into the generalization ability of learned image reconstruction with respect to deviations in the acquisition settings between training and testing. It also provides an outlook for the potential of transfer learning to fine-tune trainings to a particular target application using only a small number of training cases.

Patient-specific 3D printed and augmented reality kidney and prostate cancer models: impact on patient education

Abstract: Patient-specific 3D models are being used increasingly in medicine for many applications including surgical planning, procedure rehearsal, trainee education, and patient education. To date, experiences on the use of 3D models to facilitate patient understanding of their disease and surgical plan are limited. The purpose of this study was to investigate in the context of renal and prostate cancer the impact of using 3D printed and augmented reality models for patient education. Patients with MRI-visible prostate cancer undergoing either robotic assisted radical prostatectomy or focal ablative therapy or patients with renal masses undergoing partial nephrectomy were prospectively enrolled in this IRB approved study (n = 200). Patients underwent routine clinical imaging protocols and were randomized to receive pre-operative planning with imaging alone or imaging plus a patient-specific 3D model which was either 3D printed, visualized in AR, or viewed in 3D on a 2D computer monitor. 3D uro-oncologic models were created from the medical imaging data. A 5-point Likert scale survey was administered to patients prior to the surgical procedure to determine understanding of the cancer and treatment plan. If randomized to receive a pre-operative 3D model, the survey was completed twice, before and after viewing the 3D model. In addition, the cohort that received 3D models completed additional questions to compare usefulness of the different forms of visualization of the 3D models. Survey responses for each of the 3D model groups were compared using the Mann-Whitney and Wilcoxan rank-sum tests. All 200 patients completed the survey after reviewing their cases with their surgeons using imaging only. 127 patients completed the 5-point Likert scale survey regarding understanding of disease and surgical procedure twice, once with imaging and again after reviewing imaging plus a 3D model. Patients had a greater understanding using 3D printed models versus imaging for all measures including comprehension of disease, cancer size, cancer location, treatment plan, and the comfort level regarding the treatment plan (range 4.60–4.78/5 vs. 4.06–4.49/5, p < 0.05). All types of patient-specific 3D models were reported to be valuable for patient education. Out of the three advanced imaging methods, the 3D printed models helped patients to have the greatest understanding of their anatomy, disease, tumor characteristics, and surgical procedure.

Value of MRI in medicine : More than just another test?

Abstract: There is increasing scrutiny from healthcare organizations towards the utility and associated costs of imaging. MRI has traditionally been used as a high-end modality, and although shown extremely important for many types of clinical scenarios, it has been suggested as too expensive by some. This editorial will try and explain how value should be addressed and gives some insights and practical examples of how value of MRI can be increased. It requires a global effort to increase accessibility, value for money, and impact on patient management. We hope this editorial sheds some light and gives some indications of where the field may wish to address some of its research to proactively demonstrate the value of MRI. Level of Evidence: 5 Technical Efficacy: Stage 5 J. Magn. Reson. Imaging 2019;49:e14-e25.

Magnetization Transfer in Magnetic Resonance Fingerprinting

Abstract: Purpose: To study the effects of magnetization transfer (MT, in which a semisolid spin pool interacts with the free pool), in the context of magnetic resonance fingerprinting (MRF). Methods: Simulations and phantom experiments were performed to study the impact of MT on the MRF signal and its potential influence on T1 and T2 estimation. Subsequently, an MRF sequence implementing off-resonance MT pulses and a dictionary with an MT dimension by incorporating a two-pool model were used to estimate the fractional pool size in addition to the B1+, T1, and T2 values. The proposed method was evaluated in the human brain. Results: Simulations and phantom experiments showed that an MRF signal obtained from a cross-linked bovine serum sample is influenced by MT. Using a dictionary based on an MT model, a better match between simulations and acquired MR signals can be obtained (NRMSE 1.3% versus 4.7%). Adding off-resonance MT pulses can improve the differentiation of MT from T1 and T2. In-vivo results showed that MT affects the MRF signals from white matter (fractional pool-size ~16%) and gray matter (fractional pool-size ~10%). Furthermore, longer T1 (~1060 ms versus ~860 ms) and T2 values (~47 ms versus ~35 ms) can be observed in white matter if MT is accounted for. Conclusion: Our experiments demonstrated a potential influence of MT on the quantification of T1 and T2 with MRF. A model that encompasses MT effects can improve the accuracy of estimated relaxation parameters and allows quantification of the fractional pool size.

Deep Learning Methods for Parallel Magnetic Resonance Image Reconstruction

Abstract: Following the success of deep learning in a wide range of applications, neural network-based machine learning techniques have received interest as a means of accelerating magnetic resonance imaging (MRI). A number of ideas inspired by deep learning techniques from computer vision and image processing have been successfully applied to non-linear image reconstruction in the spirit of compressed sensing for both low dose computed tomography and accelerated MRI. The additional integration of multi-coil information to recover missing k-space lines in the MRI reconstruction process, is still studied less frequently, even though it is the de-facto standard for currently used accelerated MR acquisitions. This manuscript provides an overview of the recent machine learning approaches that have been proposed specifically for improving parallel imaging. A general background introduction to parallel MRI is given that is structured around the classical view of image space and k-space based methods. Both linear and non-linear methods are covered, followed by a discussion of recent efforts to further improve parallel imaging using machine learning, and specifically using artificial neural networks. Image-domain based techniques that introduce improved regularizers are covered as well as k-space based methods, where the focus is on better interpolation strategies using neural networks. Issues and open problems are discussed as well as recent efforts for producing open datasets and benchmarks for the community.

Simultaneous Evaluation of Lung Anatomy and Ventilation Using 4D Respiratory-Motion-Resolved Ultrashort Echo Time Sparse MRI.

Abstract: BACKGROUND Computed tomography (CT) and spirometry are the current standard methods for assessing lung anatomy and pulmonary ventilation, respectively. However, CT provides limited ventilation information and spirometry only provides global measures of lung ventilation. Thus, a method that can enable simultaneous examination of lung anatomy and ventilation is of clinical interest. PURPOSE To develop and test a 4D respiratory-resolved sparse lung MRI (XD-UTE: eXtra-Dimensional Ultrashort TE imaging) approach for simultaneous evaluation of lung anatomy and pulmonary ventilation. STUDY TYPE Prospective. POPULATION In all, 23 subjects (11 volunteers and 12 patients, mean age = 63.6 ± 8.4). FIELD STRENGTH/SEQUENCE 3T MR; a prototype 3D golden-angle radial UTE sequence, a Cartesian breath-hold volumetric-interpolated examination (BH-VIBE) sequence. ASSESSMENT All subjects were scanned using the 3D golden-angle radial UTE sequence during normal breathing. Ten subjects underwent an additional scan during alternating normal and deep breathing. Respiratory-motion-resolved sparse reconstruction was performed for all the acquired data to generate dynamic normal-breathing or deep-breathing image series. For comparison, BH-VIBE was performed in 12 subjects. Lung images were visually scored by three experienced chest radiologists and were analyzed by two observers who segmented the left and right lung to derive ventilation parameters in comparison with spirometry. STATISTICAL TESTS Nonparametric paired two-tailed Wilcoxon signed-rank test; intraclass correlation coefficient, Pearson correlation coefficient. RESULTS XD-UTE achieved significantly improved image quality compared both with Cartesian BH-VIBE and radial reconstruction without motion compensation (P < 0.05). The global ventilation parameters (a sum of the left and right lung measures) were in good correlation with spirometry in the same subjects (correlation coefficient = 0.724). There were excellent correlations between the results obtained by two observers (intraclass correlation coefficient ranged from 0.8855-0.9995). DATA CONCLUSION Simultaneous evaluation of lung anatomy and ventilation using XD-UTE is demonstrated, which have shown good potential for improved diagnosis and management of patients with heterogeneous lung diseases. LEVEL OF EVIDENCE 2 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2019;49:411-422.

GrappaNet: Combining Parallel Imaging with Deep Learning for Multi-Coil MRI Reconstruction.

Abstract: Magnetic Resonance Image (MRI) acquisition is an inherently slow process which has spurred the development of two different acceleration methods: acquiring multiple correlated samples simultaneously (parallel imaging) and acquiring fewer samples than necessary for traditional signal processing methods (compressed sensing). Both methods provide complementary approaches to accelerating the speed of MRI acquisition. In this paper, we present a novel method to integrate traditional parallel imaging methods into deep neural networks that is able to generate high quality reconstructions even for high acceleration factors. The proposed method, called GrappaNet, performs progressive reconstruction by first mapping the reconstruction problem to a simpler one that can be solved by a traditional parallel imaging methods using a neural network, followed by an application of a parallel imaging method, and finally fine-tuning the output with another neural network. The entire network can be trained end-to-end. We present experimental results on the recently released fastMRI dataset and show that GrappaNet can generate higher quality reconstructions than competing methods for both $4\times$ and $8\times$ acceleration.

Optimized quantification of spin relaxation times in the hybrid state.

Abstract: Purpose The optimization and analysis of spin ensemble trajectories in the hybrid state-a state in which the direction of the magnetization adiabatically follows the steady state while the magnitude remains in a transient state. Methods Numerical optimizations were performed to find spin ensemble trajectories that minimize the Cramer-Rao bound for T 1 -encoding, T 2 -encoding, and their weighted sum, respectively, followed by a comparison between the Cramer-Rao bounds obtained with our optimized spin-trajectories, Look-Locker sequences, and multi-spin-echo methods. Finally, we experimentally tested our optimized spin trajectories with in vivo scans of the human brain. Results After a nonrecurring inversion segment on the southern half of the Bloch sphere, all optimized spin trajectories pursue repetitive loops on the northern hemisphere in which the beginning of the first and the end of the last loop deviate from the others. The numerical results obtained in this work align well with intuitive insights gleaned directly from the governing equation. Our results suggest that hybrid-state sequences outperform traditional methods. Moreover, hybrid-state sequences that balance T 1 - and T 2 -encoding still result in near optimal signal-to-noise efficiency for each relaxation time. Thus, the second parameter can be encoded at virtually no extra cost. Conclusions We provided new insights into the optimal encoding processes of spin relaxation times in order to guide the design of robust and efficient pulse sequences. We found that joint acquisitions of T 1 and T 2 in the hybrid state are substantially more efficient than sequential encoding techniques.

Training a Neural Network for Gibbs and Noise Removal in Diffusion MRI

Abstract: We develop and evaluate a neural network-based method for Gibbs artifact and noise removal. A convolutional neural network (CNN) was designed for artifact removal in diffusion-weighted imaging data. Two implementations were considered: one for magnitude images and one for complex images. Both models were based on the same encoder-decoder structure and were trained by simulating MRI acquisitions on synthetic non-MRI images. Both machine learning methods were able to mitigate artifacts in diffusion-weighted images and diffusion parameter maps. The CNN for complex images was also able to reduce artifacts in partial Fourier acquisitions. The proposed CNNs extend the ability of artifact correction in diffusion MRI. The machine learning method described here can be applied on each imaging slice independently, allowing it to be used flexibly in clinical applications.

Hybrid-state free precession in nuclear magnetic resonance.

Abstract: The dynamics of large spin-1/2 ensembles are commonly described by the Bloch equation, which is characterized by the magnetization's non-linear response to the driving magnetic field. Consequently, most magnetic field variations result in non-intuitive spin dynamics, which are sensitive to small calibration errors. Although simplistic field variations result in robust spin dynamics, they do not explore the richness of the system's phase space. Here, we identify adiabaticity conditions that span a large experiment design space with tractable dynamics. All dynamics are trapped in a one-dimensional subspace, namely in the magnetization's absolute value, which is in a transient state, while its direction adiabatically follows the steady state. In this hybrid state, the polar angle is the effective drive of the spin dynamics. As an example, we optimize this drive for robust and efficient quantification of spin relaxation times and utilize it for magnetic resonance imaging of the human brain.

A New Method for Cartilage Evaluation in Femoroacetabular Impingement Using Quantitative T2 Magnetic Resonance Imaging: Preliminary Validation against Arthroscopic Findings.

Abstract: ObjectiveThe outcome of arthroscopic treatment for femoroacetabular impingement (FAI) depends on the preoperative status of the hip cartilage. Quantitative T2 can detect early biochemical cartilage...

Size‐adaptable “Trellis” structure for tailored MRI coil arrays

Abstract: We present a novel, geometrically adjustable, receive coil array whose diameter can be tailored to the subject in order to maximize sensitivity for a range of body sizes. A key mechanical feature of the size-adaptable receive array is its trellis structure that was motivated by similar structures found in gardening and fencing. Our implementation is a cylindrical trellis that features encircling, diagonally interleaved slats, which are linked together at intersecting points. The ensemble allows expansion or contraction to be controlled with the angle between the slats. This mechanical frame provides a base for radiofrequency coils wherein approximately constant overlap, and therefore coupling between adjacent elements, is maintained when the trellis is expanded or contracted. We demonstrate 2 trellis coil concepts for imaging lower extremity at 3T: a single-row 8-channel array built on a trellis support structure and a multirow 24-channel array in which the coil elements themselves form the trellis structure. We show that the adjustable trellis array can accommodate a range of subject sizes with robust signal-to-noise ratio, loading, and coupling. The trellis coil concept enables an array of surface coils to expand and contract with negligible effect on tuning, matching, and decoupling. This allows an encircling array to conform closely to anatomy of various sizes, which provides significant gains in signal-to-noise ratio.

Image reconstruction for interrupted-beam x-ray CT on diagnostic clinical scanners.

Abstract: Low-dose x-ray CT is a major research area with high clinical impact. Compressed sensing using view-based sparse sampling and sparsity-promoting regularization has shown promise in simulations, but these methods can be difficult to implement on diagnostic clinical CT scanners since the x-ray beam cannot be switched on and off rapidly enough. An alternative to view-based sparse sampling is interrupted-beam sparse sampling. SparseCT is a recently-proposed interrupted-beam scheme that achieves sparse sampling by blocking a portion of the beam using a multislit collimator (MSC). The use of an MSC necessitates a number of modifications to the standard compressed sensing reconstruction pipeline. In particular, we find that SparseCT reconstruction is feasible within a model-based image reconstruction framework that incorporates data fidelity weighting to consider penumbra effects and source jittering to consider the effect of partial source obstruction. Here, we present these modifications and demonstrate their application in simulations and real-world prototype scans. In simulations compared to conventional low-dose acquisitions, SparseCT is able to achieve smaller normalized root-mean square differences and higher structural similarity measures on two reduction factors. In prototype experiments, we successfully apply our reconstruction modifications and maintain image resolution at quarter-dose reduction level. The SparseCT design requires only small hardware modifications to current diagnostic clinical scanners, opening up new possibilities for CT dose reduction.

SparseCT: System concept and design of multislit collimators.

Abstract: Purpose SparseCT, an undersampling scheme for compressed sensing (CS) computed tomography (CT), has been proposed to reduce radiation dose by acquiring undersampled projection data from clinical CT scanners (Koesters et al. in, SparseCT: Interrupted-Beam Acquisition and Sparse Reconstruction for Radiation Dose Reduction; 2017). SparseCT partially blocks the x-ray beam with a multislit collimator (MSC) to perform a multidimensional undersampling along the view and detector row dimensions. SparseCT undersamples the projection data within each view and moves the MSC along the z-direction during gantry rotation to change the undersampling pattern. It enables reconstruction of images from undersampled data using CS algorithms. The purpose of this work is to design the spacing and width of the MSC slits and the MSC motion patterns based on beam separation, undersampling efficiency, and image quality. The development and testing of a SparseCT prototype with the designed MSC will be described in a following paper. Methods We chose a few initial MSC designs based on the guidance from two metrics: beam separation and undersampling efficiency. Both beam separation and undersampling efficiency were measured from numerically simulated photon distribution with MSC taken into consideration. Beam separation measures the separation between x-ray beams from consecutive slits, taking into account penumbra effects on both sides of each slit. Undersampling efficiency measures the dose-weighted similarity between penumbra undersampling and binary undersampling, in other words, the effective contribution of the incident dose to the signal to noise ratio of the projection data. We then compared the initially chosen MSC designs in terms of their reconstruction image quality. SparseCT projections were simulated from fully sampled patient projection data according to the MSC design and motion pattern, reconstructed iteratively using a sparsity-enforcing penalized weighted least squares cost function with ordered subsets/momentum algorithm, and compared visually and quantitatively. Results Simulated photon distributions indicate that the size of the penumbra is dominated by the size of the focal spot. Therefore, a wider MSC slit and a smaller focal spot lead to increased beam separation and undersampling efficiency. For fourfold undersampling with a 1.2 mm focal spot, a minimum MSC slit width of three detector rows (projected to the detector surface) is needed for beam separation; for threefold undersampling, a minimum slit width of four detector rows is needed. Simulations of SparseCT projection and reconstruction indicate that the motion pattern of the MSC does not have a visible impact on image quality. An MSC slit width of three or four detector rows yields similar image quality. Conclusion The MSC is the key component of the SparseCT method. Simulations of MSC designs incorporating x-ray beam penumbra effects showed that for threefold and fourfold dose reductions, an MSC slit width of four detector rows provided reasonable beam separation, undersampling efficiency, and image quality.

Disentangling the effects of high permittivity materials on signal optimization and sample noise reduction via ideal current patterns

Abstract: PURPOSE To investigate how high-permittivity materials (HPMs) can improve SNR when placed between MR detectors and the imaged body. METHODS We used a simulation framework based on dyadic Green's functions to calculate the electromagnetic field inside a uniform dielectric sphere at 7 Tesla, with and without a surrounding layer of HPM. SNR-optimizing (ideal) current patterns were expressed as the sum of signal-optimizing (signal-only) current patterns and dark mode current patterns that minimize sample noise while contributing nothing to signal. We investigated how HPM affects the shape and amplitude of these current patterns, sample noise, and array SNR. RESULTS Ideal and signal-only current patterns were identical for a central voxel. HPMs introduced a phase shift into these patterns, compensating for signal propagation delay in the HPMs. For an intermediate location within the sphere, dark mode current patterns were present and illustrated the mechanisms by which HPMs can reduce sample noise. High-amplitude signal-only current patterns were observed for HPM configurations that shield the electromagnetic field from the sample. For coil arrays, these configurations corresponded to poor SNR in deep regions but resulted in large SNR gains near the surface due to enhanced fields in the vicinity of the HPM. For very high relative permittivity values, HPM thicknesses corresponding to even multiples of λ/4 resulted in coil SNR gains throughout the sample. CONCLUSION HPMs affect both signal sensitivity and sample noise. Lower amplitude signal-only optimal currents corresponded to higher array SNR performance and could guide the design of coils integrated with HPM.

Population net benefit of prostate MRI with high spatiotemporal resolution contrast-enhanced imaging: A decision curve analysis.

Abstract: Background The value of dynamic contrast-enhanced (DCE) sequences in prostate MRI compared with noncontrast MRI is controversial. Purpose To evaluate the population net benefit of risk stratification using DCE-MRI for detection of high-grade prostate cancer (HGPCA), with or without high spatiotemporal resolution DCE imaging. Study type Decision curve analysis. Population Previously published patient studies on MRI for HGPCA detection, one using DCE with golden-angle radial sparse parallel (GRASP) images and the other using standard DCE-MRI. Field strength/sequence GRASP or standard DCE-MRI at 3 T. Assessment Each study reported the proportion of lesions with HGPCA in each Prostate Imaging Reporting and Data System version 2 (PI-RADS v2) category (1-5), before and after reclassification of peripheral zone lesions from PI-RADS 3-4 based on contrast-enhanced images. This additional risk stratifying information was translated to population net benefit, when biopsy was hypothetically performed for: all lesions, no lesions, PI-RADS ≥3 (using NC-MRI), and PI-RADS ≥4 on DCE. Statistical tests Decision curve analysis was performed for both GRASP and standard DCE-MRI data, translating the avoidance of unnecessary biopsies and detection of HGPCA to population net benefit. We standardized net benefit values for HGPCA prevalence and graphically summarized the comparative net benefit of biopsy strategies. Results For a clinically relevant range of risk thresholds for HGPCA (>11%), GRASP DCE-MRI with biopsy of PI-RADS ≥4 lesions provided the highest net benefit, while biopsy of PI-RADS ≥3 lesions provided highest net benefit at low personal risk thresholds (2-11%). In the same range of risk thresholds using standard DCE-MRI, the optimal strategy was biopsy for all lesions (0-15% risk threshold) or PI-RADS ≥3 on NC-MRI (16-33% risk threshold). Data conclusion GRASP DCE-MRI may potentially enable biopsy of PI-RADS ≥4 lesions, providing relatively preserved detection of HGPCA and avoidance of unnecessary biopsies compared with biopsy of all PI-RADS ≥3 lesions. J. Magn. Reson. Imaging 2019;49:1400-1408.

Resonate: Reflections and recommendations on implicit biases within the ISMRM

Abstract: Why Do Diversity, Equity, and Inclusiveness Matter to ISMRM? Science undeniably evolves through collaboration, transparency, and inclusiveness. Unconscious attitudes or stereotypes based on culture, personal experiences, or institutional influences, ie, implicit bias, can either positively or negatively inform our understanding, actions, and decisions. The negative effects of implicit bias threaten the intrinsic worth of the scientific method and minimize progress. There is growing evidence that diverse research teams are more productive and make better group decisions and the use of diverse study populations leads to more impactful science. As an example, LeWinn et al recently challenged the—often implicit— assumption in population selection for neuroimaging studies that basic neural functions are not influenced by sample characteristics. LeWinn et al used >1000 samples from the Pediatric Imaging, Neurocognition and Genetics study to show that age-related changes in brain structure are dependent on the composition of the sample, thereby highlighting the need for study populations to reflect target populations of interest to ensure generalizability of the study outcomes. The ISMRM, like others in Science, Technology, Engineering, Maths (STEM) fields, face challenges because of implicit bias. For example, although the overall membership of women in the ISMRM is slowly increasing (growing from 21% in 2008 to 27% in 2017), there remains a significant disparity in the representation of women among student members (35% female) when compared to full members (21% female). This gap between young and senior female scientists is persistent and consistent across STEM fields. Moss-Racusin et al suggest that interventions addressing gender bias might advance women’s participation in STEM fields. Consistent with this body of literature, in 2013 the first annual \"Women in MR forum\" was organized at the annual meeting of the ISMRM. This event, arguably, marks the start of the society’s efforts to openly address implicit bias towards gender. This year, we (the members and the society’s leadership) extended the conversation at our annual scientific meeting by highlighting the further range of implicit biases that affect our science in magnetic resonance imaging (MRI). Resonate: A Community-Wide Conversation on Implicit Bias and Equity in ISMRM Various events organized during the 26 Annual Meeting of the ISMRM were designed to identify and address implicit biases that the ISMRM membership are facing today. These events included 1) an inaugural Presidential Lecture delivered by Professor Curt Rice, summarizing current research on diversity in research organizations; 2) a Member-Initiated Symposium entitled Resonate: A Discussion on Social Biases Within the ISMRM; 3) Women in MR forum focusing on gender bias; and 4) an informal secret session on Hacks for Dealing With Bias. Implicit-bias-related issues raised by the ISMRM membership across these events include gender equality, international diversity, LGBTQA in STEM, accessibility for people with disabilities, and other barriers to member participation, such as a lack of childcare facilities at meetings and workshops. Certain demographic examples indicative of bias within the ISMRM were discussed—for example, the fact that, of 77 ISMRM Gold Medal awardees, only four have been women, and only one has been based outside North America or Europe.