KomaMRI.jl: An open-source framework for general MRI simulations with GPU acceleration

PURPOSE

To develop an open-source, high-performance, easy-to-use, extensible, cross-platform, and general MRI simulation framework (Koma).

METHODS

Koma was developed using the Julia programming language. Like other MRI simulators, it solves the Bloch equations with CPU and GPU parallelization. The inputs are the scanner parameters, the phantom, and the pulse sequence that is Pulseq-compatible. The raw data is stored in the ISMRMRD format. For the reconstruction, MRIReco.jl is used. A graphical user interface utilizing web technologies was also designed. Two types of experiments were performed: one to compare the quality of the results and the execution speed, and the second to compare its usability. Finally, the use of Koma in quantitative imaging was demonstrated by simulating Magnetic Resonance Fingerprinting (MRF) acquisitions.

RESULTS

Koma was compared to two well-known open-source MRI simulators, JEMRIS and MRiLab. Highly accurate results (with mean absolute differences below 0.1% compared to JEMRIS) and better GPU performance than MRiLab were demonstrated. In an experiment with students, Koma was proved to be easy to use, eight times faster on personal computers than JEMRIS, and 65% of test subjects recommended it. The potential for designing acquisition and reconstruction techniques was also shown through the simulation of MRF acquisitions, with conclusions that agree with the literature.

CONCLUSIONS

Koma's speed and flexibility have the potential to make simulations more accessible for education and research. Koma is expected to be used for designing and testing novel pulse sequences before implementing them in the scanner with Pulseq files, and for creating synthetic data to train machine learning models.

Simulated and experimental phantom data for multi-center quality assurance of quantitative susceptibility maps at 3 T, 7 T and 9.4 T

PURPOSE

To develop methods for quality assurance of quantitative susceptibility mapping (QSM) using MRI at different magnetic field strengths, and scanners, using different MR-sequence protocols, and post-processing pipelines.

METHODS

We built a custom phantom based on iron in two forms: homogeneous susceptibility ('free iron') and with fine-scaled variations in susceptibility ('clustered iron') at different iron concentrations. The phantom was measured at 3.0 T (two scanners), 7.0 T and 9.4 T using multi-echo, gradient echo acquisition sequences. A digital phantom analogue to the iron-phantom, tailored to obtain similar results as in experimentation was developed, with similar geometry and susceptibility values. Morphology enabled dipole inversion was applied to the phase images to obtain QSM for experimental and simulated data using the MEDI + 0 approach for background regularization.

RESULTS

Across all scanners, QSM-values showed a linear increase with iron concentrations. The QSM-relaxivity was 0.231 ± 0.047 ppm/mM for free and 0.054 ± 0.013 ppm/mM for clustered iron, with adjusted determination coefficients (DoC) ≥ 0.87. Similarly, the simulations yielded linear increases (DoC ≥ 0.99). In both the experimental and digital phantoms, the estimated molar susceptibility was lower with clustered iron, because clustering led to highly localized field effects.

CONCLUSION

Our iron phantom can be used to evaluate the capability of QSM to detect local variations in susceptibility across different field strengths, when using different MR-sequence protocols. The devised simulation method captures the effect of iron clustering in QSM as seen experimentally and could be used in the future to optimize QSM processing pipelines and achieve higher accuracy for local field effects, as also seen in Alzheimer's beta-amyloid plaques.

Receiver phase alignment using fitted SVD derived sensitivities from routine prescans

Magnetic resonance imaging radio frequency arrays are composed of multiple receive coils that have their signals combined to form an image. Combination requires an estimate of the radio frequency coil sensitivities to align signal phases and prevent destructive interference. At lower fields this can be accomplished using a uniform physical reference coil. However, at higher fields, uniform volume coils are lacking and, when available, suffer from regions of low receive sensitivity that result in poor sensitivity estimation and combination. Several approaches exist that do not require a physical reference coil but require manual intervention, specific prescans, or must be completed post-acquisition. This makes these methods impractical for large multi-volume datasets such as those collected for novel types of functional MRI or quantitative susceptibility mapping, where magnitude and phase are important. This pilot study proposes a fitted SVD method which utilizes existing combination methods to create a phase sensitive combination method targeted at large multi-volume datasets. This method uses any multi-image prescan to calculate the relative receive sensitivities using voxel-wise singular value decomposition. These relative sensitivities are fitted to the solid harmonics using an iterative least squares fitting algorithm. Fits of the relative sensitivities are used to align the phases of the receive coils and improve combination in subsequent acquisitions during the imaging session. This method is compared against existing approaches in the human brain at 7 Tesla by examining the combined data for the presence of singularities and changes in phase signal-to-noise ratio. Two additional applications of the method are also explored, using the fitted SVD method in an asymmetrical coil and in a case with subject motion. The fitted SVD method produces singularity-free images and recovers between 95-100% of the phase signal-to-noise ratio depending on the prescan data resolution. Using solid harmonic fitting to interpolate singular value decomposition derived receive sensitivities from existing prescans allows the fitted SVD method to be used on all acquisitions within a session without increasing exam duration. Our fitted SVD method is able to combine imaging datasets accurately without supervision during online reconstruction.

Effects of phase regression on high-resolution functional MRI of the primary visual cortex

High-resolution functional MRI studies have become a powerful tool to non-invasively probe the sub-millimeter functional organization of the human cortex. Advances in MR hardware, imaging techniques and sophisticated post-processing methods have allowed high resolution fMRI to be used in both the clinical and academic neurosciences. However, consensus within the community regarding the use of gradient echo (GE) or spin echo (SE) based acquisition remains largely divided. On one hand, GE provides a high temporal signal-to-noise ratio (tSNR) technique sensitive to both the macro- and micro-vascular signal while SE based methods are more specific to microvasculature but suffer from lower tSNR and specific absorption rate limitations, especially at high field and with short repetition times. Fortunately, the phase of the GE-EPI signal is sensitive to vessel size and this provides a potential avenue to reduce the macrovascular weighting of the signal (phase regression, Menon 2002). In order to determine the efficacy of this technique at high-resolution, phase regression was applied to GE-EPI timeseries and compared to SE-EPI to determine if GE-EPI's specificity to the microvascular compartment improved. To do this, functional data was collected from seven subjects on a neuro-optimized 7 T system at 800 μm isotropic resolution with both GE-EPI and SE-EPI while observing an 8 Hz contrast reversing checkerboard. Phase data from the GE-EPI was used to create a microvasculature-weighted time series (GE-EPI-PR). Anatomical imaging (MP2RAGE) was also collected to allow for surface segmentation so that the functional results could be projected onto a surface. A multi-echo gradient echo sequence was collected and used to identify venous vasculature. The GE-EPI-PR surface activation maps showed a high qualitative similarity with SE-EPI and also produced laminar activity profiles similar to SE-EPI. When the GE-EPI and GE-EPI-PR distributions were compared to SE-EPI it was shown that GE-EPI-PR had similar distribution characteristics to SE-EPI (p < 0.05) across the top 60% of cortex. Furthermore, it was shown that GE-EPI-PR has a higher contrast-to-noise ratio (0.5 ± 0.2, mean ± std. dev. across layers) than SE-EPI (0.27 ± 0.07) demonstrating the technique has higher sensitivity than SE-EPI. Taken together this evidence suggests phase regression is a useful method in low SNR studies such as high-resolution fMRI.

Sample size estimation: Current practice and considerations for original investigations in MRI technical development studies

PURPOSE

To investigate and to provide guidance for sample size selection based on the current practice in MR technical development studies in which healthy volunteers are examined.

METHODS

All original articles published in Magnetic Resonance in Medicine between 2017 and 2019 were investigated and categorized according to technique, anatomical region, and magnetic field strength. The number of examined healthy volunteers (ie, the sample size) was collected and evaluated, whereas the number of patients was not considered. Papers solely measuring patients, animals, phantoms, specimens, or studies using existing data, for example, from an open databank, or consisting only of theoretical work or simulations were excluded.

RESULTS

The median sample size of the 882 included studies was 6. There were some peaks in the sample size distribution (eg, 1, 5, and 10). In 49.9%, 82.1%, and 95.6% of the studies, the sample size was smaller or equal to 5, 10, and 20, respectively.

CONCLUSION

We observed a large variance in sample sizes reflecting the variety of studies published in Magnetic Resonance in Medicine. Therefore, it can be concluded that it is current practice to balance the need for statistical power with the demand to minimize experiments involving healthy humans, often by choosing small sample sizes between 1 and 10. Naturally, this observation does not release an investigator from ensuring that sufficient data are acquired to reach statistical conclusions.

Towards Optimising MRI Characterisation of Tissue (TOMCAT) Dataset including all Longitudinal Automatic Segmentation of Hippocampal Subfields (LASHiS) data

Seven healthy participants were scanned using a Siemens Magnetom 7 Tesla (T) whole-body research MRI scanner (Siemens Healthcare, Erlangen, Germany). The first scan session was acquired in 2016 (time point one), the second and third session in 2019 (time point two and three, respectively) with the third session acquired 45 min following the second as a scan-rescan condition. The following scans were acquired for all time points: structural T1 weighted (T1w) MP2RAGE, high in-plane resolution Turbo-Spin Echo (TSE) dedicated for hippocampus subfield segmentation. The data were used in three projects to date, for more insight see: 1) Non-linear realignment for Turbo-Spin Echo retrospective motion correction and hippocampus segmentation improvement [1] 2) Longitudinal Automatic Segmentation of Hippocampal Subfields (LASHiS) using multi-contrast MRI [2]. 3) The challenge of bias-free coil combination for quantitative susceptibility mapping at ultra-high field [3]. Data were converted from DICOM to nifti format following the Brain Imaging Data Structure (BIDS) [4]. Data were analysed for the accompanying manuscript "Longitudinal Automatic Segmentation of Hippocampal Subfields (LASHiS) using multi-contrast MRI" including test-retest reliability and longitudinal Bayesian Linear Mixed Effects (LME) modelling.

Multi-centre, multi-vendor reproducibility of 7T QSM and R2* in the human brain: Results from the UK7T study

Introduction

We present the reliability of ultra-high field T₂* MRI at 7T, as part of the UK7T Network's “Travelling Heads” study. T₂*-weighted MRI images can be processed to produce quantitative susceptibility maps (QSM) and R₂* maps. These reflect iron and myelin concentrations, which are altered in many pathophysiological processes. The relaxation parameters of human brain tissue are such that R₂* mapping and QSM show particularly strong gains in contrast-to-noise ratio at ultra-high field (7T) vs clinical field strengths (1.5–3T). We aimed to determine the inter-subject and inter-site reproducibility of QSM and R₂* mapping at 7T, in readiness for future multi-site clinical studies.

Methods

Ten healthy volunteers were scanned with harmonised single- and multi-echo T₂*-weighted gradient echo pulse sequences. Participants were scanned five times at each “home” site and once at each of four other sites. The five sites had 1× Philips, 2× Siemens Magnetom, and 2× Siemens Terra scanners. QSM and R₂* maps were computed with the Multi-Scale Dipole Inversion (MSDI) algorithm (https://github.com/fil-physics/Publication-Code). Results were assessed in relevant subcortical and cortical regions of interest (ROIs) defined manually or by the MNI152 standard space.

Results and Discussion

Mean susceptibility (χ) and R₂* values agreed broadly with literature values in all ROIs. The inter-site within-subject standard deviation was 0.001–0.005 ppm (χ) and 0.0005–0.001 ms⁻¹ (R₂*). For χ this is 2.1–4.8 fold better than 3T reports, and 1.1–3.4 fold better for R₂*. The median ICC from within- and cross-site R₂* data was 0.98 and 0.91, respectively. Multi-echo QSM had greater variability vs single-echo QSM especially in areas with large B₀ inhomogeneity such as the inferior frontal cortex. Across sites, R₂* values were more consistent than QSM in subcortical structures due to differences in B₀-shimming. On a between-subject level, our measured χ and R₂* cross-site variance is comparable to within-site variance in the literature, suggesting that it is reasonable to pool data across sites using our harmonised protocol.

Conclusion

The harmonized UK7T protocol and pipeline delivers on average a 3-fold improvement in the coefficient of reproducibility for QSM and R₂* at 7T compared to previous reports of multi-site reproducibility at 3T. These protocols are ready for use in multi-site clinical studies at 7T.

MRI phase offset correction method impacts quantitative susceptibility mapping

Individual channel ultra-high field (7T) phase images have to be phase offset corrected prior to the mapping of magnetic susceptibility of tissue. Whilst numerous methods have been proposed for gradient recalled echo MRI phase offset correction, it remains unclear how they affect quantitative magnetic susceptibility values derived from phase images. Methods already proposed either employ a single or multiple echo time MRI data. In terms of the latter, offsets can be derived using an ultra-short echo time acquisition, or by estimating the offset based on two echo points with the assumption of linear phase evolution with echo time. Our evaluation involved 32 channel multi-echo time 7T GRE (Gradient Recalled Echo) and ultra-short echo time PETRA (Pointwise Encoding Time Reduction with Radial Acquisition) MRI data collected for a susceptibility phantom and three human brains. The combined phase images generated using four established offset correction methods (two single and two multiple echo time) were analysed, followed by an assessment of quantitative susceptibility values obtained for a phantom and human brains. The effectiveness of each method in removing the offsets was shown to reduce with increased echo time, decreased signal intensity and reduced overlap in coil sensitivity profiles. Quantitative susceptibility values and how they change with echo time were found to be method specific. Phase offset correction methods based on single echo time data have a tendency to produce more accurate and less noisy quantitative susceptibility maps in comparison with methods employing multiple echo time data.

Field strength influences on gradient recalled echo MRI signal compartment frequency shifts

Different trends of echo time dependent gradient recalled echo MRI signals in different brain regions have been attributed to signal compartments in image voxels. It remains unclear how variations in gradient recalled echo MRI signals change as a function of MRI field strength, and how data processing may impact signal compartment parameters. We used two popular quantitative susceptibility mapping methods of processing raw phase images (Laplacian and path-based unwrapping with V-SHARP) and expressed values in the form of induced frequency shifts (in Hz) in six specific brain regions at 3T and 7T. We found the frequency shift curves to vary with echo time, and a good overlap between 3T and 7T mean frequency shift curves was present. However, the amount of variation across participants was greater at 3T, and we were able to obtain better compartment model fits of the signal at 7T. We also found the temporal trends in the signal and compartment frequency shifts to change with the method used to process images. The inter-participant averaged trends were consistent between 3T and 7T for each quantitative susceptibility pipeline. However, signal compartment frequency shifts generated using different pipelines may not be comparable.

Longitudinal Automatic Segmentation of Hippocampal Subfields (LASHiS) using multi-contrast MRI

The volumetric and morphometric examination of hippocampus formation subfields in a longitudinal manner using in vivo MRI could lead to more sensitive biomarkers for neuropsychiatric disorders and diseases including Alzheimer's disease, as the anatomical subregions are functionally specialised. Longitudinal processing allows for increased sensitivity due to reduced confounds of inter-subject variability and higher effect-sensitivity than cross-sectional designs. We examined the performance of a new longitudinal pipeline (Longitudinal Automatic Segmentation of Hippocampus Subfields [LASHiS]) against three freely available, published approaches. LASHiS automatically segments hippocampus formation subfields by propagating labels from cross-sectionally labelled time point scans using joint-label fusion to a non-linearly realigned 'single subject template', where image segmentation occurs free of bias to any individual time point. Our pipeline measures tissue characteristics available in in vivo high-resolution MRI scans, at both clinical (3 ​T) and ultra-high field strength (7 ​T) and differs from previous longitudinal segmentation pipelines in that it leverages multi-contrast information in the segmentation process. LASHiS produces robust and reliable automatic multi-contrast segmentations of hippocampus formation subfields, as measured by higher volume similarity coefficients and Dice coefficients for test-retest reliability and robust longitudinal Bayesian Linear Mixed Effects results at 7 ​T, while showing sound results at 3 ​T. All code for this project including the automatic pipeline is available at https://github.com/CAIsr/LASHiS.

Coil combination using linear deconvolution in k-space for phase imaging

Background

The combination of multi-channel data is a critical step for the imaging of phase and susceptibility contrast in magnetic resonance imaging (MRI). Magnitude-weighted phase combination methods often produce noise and aliasing artifacts in the magnitude images at accelerated imaging sceneries. To address this issue, an optimal coil combination method through deconvolution in k-space is proposed in this paper.

Methods

The proposed method firstly employs the sum-of-squares and phase aligning method to yield a complex reference coil image which is then used to calculate the coil sensitivity and its Fourier transform. Then, the coil k-space combining weights is computed, taking into account the truncated frequency data of coil sensitivity and the acquired k-space data. Finally, combining the coil k-space data with the acquired weights generates the k-space data of proton distribution, with which both phase and magnitude information can be obtained straightforwardly. Both phantom and in vivo imaging experiments were conducted to evaluate the performance of the proposed method.

Results

Compared with magnitude-weighted method and MCPC-C, the proposed method can alleviate the phase cancellation in coil combination, resulting in a less wrapped phase.

Conclusions

The proposed method provides an effective and efficient approach to combine multiple coil image in parallel MRI reconstruction, and has potential to benefit routine clinical practice in the future.

Cyclic Continuous Max-Flow: A Third Paradigm in Generating Local Phase Shift Maps in MRI

Sensitivity to phase deviations in MRI forms the basis of a variety of techniques, including magnetic susceptibility weighted imaging and chemical shift imaging. Current phase processing techniques fall into two families: those which process the complex image data with magnitude and phase coupled, and phase unwrapping-based techniques that first linearize the phase topology across the image. However, issues, such as low signal and the existence of phase poles, can lead both methods to experience error. Cyclic continuous max-flow (CCMF) phase processing uses primal-dual-variational optimization over a cylindrical manifold, which represent the inherent topology of phase images, increasing its robustness to these issues. CCMF represents a third distinct paradigm in phase processing, being the only technique equipped with the inherent topology of phase. CCMF is robust and efficient with at least comparable accuracy as the prior paradigms.