High-level Nipype data processing pipeline with visual illustrations of intermediate results and outputs. The inputs to single-subject processes are those of individual subjects, whereas group processes use the results from all subjects as input.

QSM template based on 45 subjects. Results were reconstructed using our QSMxT framework within our modular reconstruction and analysis ecosystem utilising the newly proposed masking technique free of anatomical assumptions. The template was generated using the minimum deformation averaging tool, volgenmodel.

Figure 1. Physiopy contributors. Contributions are recognized according to the all-contributors specification, see emoji key for details.

Figure 2. Using phys2bids to inspect the contents of physiological recording files. Choosing the ‘-info’ option in phys2bids allows users to inspect their data files before processing, providing information about recorded channels along with a graph.

Figure 2. Manufacturer-specific software implementations for Siemens (top), GE (middle) and Philips (bottom) platforms. Standard manufacturer infrastructures (Siemens IDEA/ICE, GE Python Orchestra and Philips PRIDE 2.0) are used to call the VIE module, which is a pure Python implementation installed by copying a simple standard directory structure onto a workstation. VIE results are sent to MR image databases as DICOM ‘screenshot’ images of spectral plots and tabular metabolite fit results.

Figure 3. Fitting results for a test subject data taken on a Siemens Trio MR scanner. Image resolution is 1024x1024 pixels. Spectral plots include raw data (black), baseline estimate (green) and metabolite+macromolecule+baseline total fit (red), metabolite fit values are in the table. Fitting time was approximately 90 seconds on the Siemens MRIR computer. An example of pixilation is shown in the blown-up box, top right.

Figure 1. TOP: Acquisition of post-mortem MRI, dissection photographs, and histology images. Motor and extramotor blocks were sampled differently, yielding one or two photographic intermediates (the block-face and the coronal slice images). BOTTOM: Histology-to-MRI registration pipelines. Stage-specific transformations are combined to map the high-resolution histology (0.5 mm/px) image into MRI space (0.25 mm/vx). The MRI data is resampled by cubic spline interpolation to produce a 2-D image.

Figure 5. Representative side-by-side comparison of histology and MRI data for a set of extramotor blocks. End-to-end registrations have uniform submillimetre accuracy in all tested anatomical regions. Grid spacing: 5 mm (10 mm for the thalamus block).

Fig 1: Open-source sequence development and image reconstruction workflow. The Pulseq sequence file and the protocol file are created with Python and exported to the scanner. The sequence can be selected in the scanner GUI and is executed by the interpreter. Acquired data is converted to the ISMRMRD streaming format in real time by the FIRE framework. Before the online reconstruction with the BART toolbox, protocol information is inserted from the protocol file. After reconstruction, images are automatically sent back to the scanner GUI using the ISMRMRD image format.

Fig 2: The reconstruction pipeline starts with continuous data streaming of the raw data to the ISMRMRD format. Header information from the protocol file is inserted into the raw data. Afterwards protocol information for each acquisition is transfered successively. This is accompanied by a trajectory prediction with the GIRF. Acquisitions are collected until data from a whole slice is available. The data is sorted and the reconstruction is performed. If the acquisition is accelerated, sensitivity maps are calculated from a reference scan in a preceding step.

Figure 1: Schematic description of the contrast mapping approach, through which a realistic “T1w-like” BigBrain image was obtained from a lower-resolution T1w image and the high-resolution BigBrain image. This was achieved by estimating and applying region-specific transfer functions between the datasets (blue module), combined with a model for partial volume contributions to properly reproduce region borders (orange).

Figure 2: Close-up examples of brain regions illustrating the performance of the super-resolution approach. This super-resolved (SR) image was reconstructed using 4 different orientations and λ=3.

Figure 2: In vivo results obtained from one healthy subject, including the estimated MWF maps by different schemes, T1W images, and B0 field maps. As can be seen, the MWF maps from the proposed method are of much higher quality than those produced by the other methods, especially in the regions where B0 inhomogeneities are large (e.g. the region marked by blue arrows).

Figure 4: Representative results obtained from multiple healthy subjects. The MWF maps produced by the proposed method are consistent among different subjects.

Figure 3. Left: ACR phantom images for 6 b-values, each of them reconstructed from a set of six interleaved SPEN data acquisitions following the processing described in Ref. 8. Center: Idem but from a set of undersampled data, using only one of the interleaved scans per b-value and following Eq. (2). Right: ADC maps calculated from b=0 and 800 s/mm² DW images by both methods. For the subspace LLR reconstruction diffusion was modeled using K=2 eigenvectors after SVD of monoexponential functions modeling signal attenuation as function of b for ADCs in the 0.1-5x10-3 s/mm² range.

Figure 5. Comparing different strategies for processing a 6-b-value (0-1000 s/mm²) SPEN experiment. (A) Usual DWI reconstruction, illustrating two b-valued images and the resulting ADC map.⁸ (B) Same but with constrained DW subspace LLR reconstruction on the full set. (C) Idem as in B, but for data with a fully sampled b=0 image and only one shot for the other b’s. (D) Idem as C, but a single shot also used for the b=0 image reconstruction. (E) Same as D, but using reconstruction in Ref. 10. Data in B-E were obtained undersampling acquisition of 6 segments at 1.2x1.2x3 mm³ resolution.

Figure 1: Overview of modules and data structures in JET. a) Image representation of the JET modules in sequential order and its data structures that exist to operate each module. b) Sequential order of spectrum registration correction within the third module. The main function of JET is operated in the spectrum registration module where parameter corrections will be performed on the incoming data to quantify metabolite concentration.

Figure 2: Spectrum registration correction using JET and its performance using error analysis for different signal noise ratios. a) Spectrum registration correction in sequential order of the spectras. It can be observed that after each correction, the DIFF spectra improves. b) Bar graph representation of frequency estimation error for different SNRs. c) Bar graph representation of phase estimation error for different SNRs. It can be observed that JET reduces both parameter errors with greater SNR.

Simulation pipeline of HASTE acquisitions from segmented high-resolution anatomical images of the fetal brain. This framework offers a great flexibility in the choice of the sequence parameters but also other settings such as the age of the fetus, the SNR of the acquisitions, and the amplitude of fetal motion.

Visual inspection and comparison between simulated HASTE images and clinical acquisitions at three different gestational ages (26, 30 and 32 weeks). Various levels of fetal movement can be observed. Arrows point out typical out-of-plane motion patterns.

Figure 1: Training and Inference pipelines: 2D CNN predicts the probably of QC failure for each 2D slice. At inference time, the mean QC score for each slice is used for the volume-wise prediction. In addition, reorientations and affine transformations are used as test-time augmentations. Robustness can be ensured by outputting the penultimate layer from the 2D CNN and comparing it to the nearest class-conditional Gaussian distribution of the training data.

Figure 4: Performance evaluation for both the validation and test sets. From left to right: confusion matrices, precision recall curves, example test-set images with QC predictions and average radiologist scores. Performance improvements using test-time augmentation are shown in the precision recall curves as an increase in average precision from 0.75 to 0.8 (validation set) and 0.69 to 0.87 (test set) (abbreviations: w/o aug - without test-time augmentation, w/ aug = with test-time augmentation).

Fig 2. The sagittal view of rigid registration results based on femoral masks at following time points. The original MR images at following time points are shown in first row. The results of rigid registration and the blend of result and mask are shown in second and third rows. The mask shown is the femoral mask of baseline.

Fig 3. The transverse view of rigid registration results based on femoral masks at following time points. We also draw the popliteal artery mask of baseline on the registration results to confirm that as long as the longitudinal registration is good, there is only a small deviation of the vessel centerline.

Figure 2: Box plots of the mean prostate intensities before and after normalization for all scanners and the entire test dataset. The red boxes are before normalization, and green are after. The dashed lines correspond to the mean fat and muscle intensity in each cohort, that have been aligned through scaling.

Table 3: The range and IQR in the scaled mean prostate intensities before and after normalization for the entire cohort. The variation in pre-normalization ranges comes from the exclusion of cases where the ROI detectors failed - leading to a slightly different dataset for each method.

White matter spatial entropy mapping. qVision processing pipeline for spatial entropy mapping, utilizing principles of MS-qMRI mapping and Synthetic MRI.

Example MS-qMRI maps of A) T1, B) T2, and C) PD for a 15-year-old female. The corresponding T1, T2, and PD histograms are provided.

Figure 1 The developed QC algorithm consists of 5 steps: 1) estimation of diffusion scalar maps; 2) normalisation of scalar maps to MNI space by TBSS procedure; 3) estimation of SSIM and skeleton-averaged metrics for each subject; 4) application of k-means-derived distances for one cluster centroid; 5) data filtration using the density-based spatial clusterisation

Figure 2 Examples of detected outliers appeared in the data. Mean kurtosis (MK) map presents problems with metric estimation. Mean diffusion (MD) map presents an anatomical specificity of the subject. Axial krtosis (AK) map presents the problem with a pair of slices misestimated by the eddy/Matlab script. Axial extraaxonal diffusivity (axEAD) map presents the problem with a flat contrast along the computations.

(A) Left cochlea template, with insert showing the 3D rendering of the inner ear specifically. (B) Slice of the spherical cochlea MRI region of one participant with manual segmentation overlaid. (C-E) 3D rendering of the manually segmented cochlea subregions and modiolus (red=cochlear basal turn, green=cochlear mid-apical turn, blue=modiolus) in different views.

Table showing automatic segmentation agreement and inter-rater segmentation agreement.

Figure 1. Magnetic Resonance Imaging (MRI) data flow in the EPAD study

Figure 4. Association of global and local tbss values with age and amyloid status. A) FA skeleton as computed in the tbss pipeline; B) Skeletonised white matter atlas used to extract local FA values; C) Association of global and regional FA values with age and amyloid status. Abbreviations: FA=Fractional Anisotropy; tbss = tract-based spatial statistics

Figure 1: Flowchart of the proposed algorithm for automated identification of spectral characteristics.

Figure 3: Box-and-whisker plots of the frequency difference between water and (main) fat peak (Delta Fat Peak) for the complete imaging volume proton spectra from the knee (number of subjects: n=9), breast (n=20), and ankle (n=11) regions. Box-and-whisker plots of the frequency difference between the water peak and the “junction” frequency (Delta Junction) are also shown. The results show the impact of field homogeneity on the spectra in different anatomical regions.

Fig. 1: Block diagram of the proposed Bayesian registration system. μ and σ are the mean and standard deviation of the deformation.

Table 1: Registration evaluation results. The added noise levels are denoted by σ and α, respectively. The mean and std, calculated over the test set, are presented for three different anatomical structures (from top to bottom: R inferior frontal gyrus, L precentral gyrus, L lateral orbitofrontal gyrus).

Figure 2. Mid-sagittal slices at four selected time frames from the reference subject and one test subject (before and after time alignment) during the pronunciation of “hamper”.

Figure 1. Wave form matching of two test subjects to the target reference subject.

Figure 1. Variation in WM Parcellations. JHU: Forceps minor (CC, ROI #9), left inferior longitudinal fascicle (ILF, ROI #12) and superior longitudinal fascicle (SLF, ROI #14); TractSeg: Rostrum (CC, ROI#5), left inferior longitudinal fascicle (ILF, ROI #26) and superior longitudinal fascicle III (SLF, ROI #39). FA and MD were lowest for single-shell acquisition (60x1). Volumes were consistent for JHU, but varied for TractSeg. Note: Atlases do not exactly segment the same volumes.

Table 1. Coefficient of Variation of ROI Means Within and Across Acquisition Schemes. All CV values displayed as mean [min, max] in %.

Plot of the number of veins counted as a function of the regularization parameter for both subjects. Data for the clinical protocol (horizontal blue dotted line) is compared to that obtained for three CS accelerations factors (4.6, 7.2 and 8.8).

Effect of increasing CS acceleration factor on the image quality, comparing: (a) clinical protocol, (b) 4.6, (c) 5.5, (d) 6.3, (e) 7.2 and (f) 8.8. SSI values, comparing to the clinical protocol in (a), were 0.45, 0.46, 0.46, 0.47, 0.46, respectively. Some residual aliasing is evident in the CS-reconstructed images, but nevertheless good overall anatomical visibility is maintained up to an acceleration factor of 7.2.

Figure 4: Comparison of coronal maximum intensity projection over the basal ganglia of the original ToF volume, vesselness-filtering and segmentation output for the proposed OMELETTE and benchmark pipeline. Additionally, the semi-automatically segmented reference is shown. OMELETTE enables better segmentation of the lenticulostriate arteries than the benchmark and reference segmentation. Note that all images were normalized.

Figure 2: Sensitivity, specificity, and Dice coefficient of OMELETTE and benchmark pipeline for 20 high resolution, publicly available ToF angiographies acquired at 7T. A semi-automatic segmentation (with ilastik) was used as a reference.

Figure 2: Exemplar coronal slices in a single mouse. Top row: MDT template (healthy template) in native space. Second row: Reference image. MDT transformed to reference image using multicomponent-ANTs (third row) and single component ANTs (last row) techniques. Multicomponent ANTs matched neuro-anatomical landmarks between MDT and individual mouse brain more robustly, with a slight overestimation of ipsilesional hippocampal boundaries.

Figure 1: A) Ipsilesional ventricle contrast enhancement. A median filter is applied to stroke images. Ipsilesional ventricle intensities are upscaled 2.5x to reconstruct an enhanced image. B) MC-registration framework. Input (MDT) and reference images are affinely registered. Images are non-linearly registered using a weighted 3-component SyN transformation. Component1: cross-correlation (CC) metric (radius:r₁), between MDT and reference images. Components2-3: CC metrics with radii r₂ and r₃, between median-filtered MDT and ventricle enhanced reference image.

Figure 1: The proposed framework consisting of synthesis and segmentation components. (a) Synthesis: The input images in cohort A are translated to synthetic images and reconstructed back to the input domain for adversarial training. (b) Segmentation: the sMRI is generated from the input CT in cohort B to provide dual-channel input for MWUNet training. In MWUNet, DWT leads to four-fold increase in the receptive field whilst reducing the dimensionality, and IWT allows loss-less feature map reconstruction.

Figure 2: The synthetic T₂W MRI generated from the test patients in cohort A through unpaired Cycle-GAN training. Whilst the texture and contrast were realistically predicted for structures with fixed geometries, soft-tissues with large variabilities in training images were challenging to reproduce in test sMRI. White arrow: rectum and bowel. Cyan arrow: muscle lining (fascia). Yellow arrow: examples of contrast disparity in femur and sacrum.

Figure 1 Flow chart of the proposed algorithms

Figure 2 Automated scan planes by using propose method

Fig3. Synthetic T1w post contrast derived from QPM and the pHe maps

Fig2. Mean values of R1subtraction, CM, relaxivity, and pHe

Figure 1. Respiratory cycle histogram created from respiration belt trace of chest position, plotted for each scan. Histograms created from chest position data during breath-hold task scans (top row) and resting-state scans (bottom row). In breath-hold scans, extra peaks at lower bins are due to prolonged low chest position during end-expiratory breath holds. Long tails at higher bins are due to large inhalations during recovery breaths post-breath hold.

Figure 2. Maps of partial R² values from adding a chest position regressor compared to RETROICOR alone during breath-hold scans. Three cervical spinal cord transverse slices are shown at approximately the same levels for each subject. Example anatomical and functional scans of the spinal cord are shown in the rightmost column for context.

Figure 2. Cases with the highest (top) and the lowest (bottom) performance. Hausdorff map is represented as a heat map in which the regions with intense yellows represent high errors and dark or black represent errors close to or equal to zero.

Figure 1. Average performance of the method during the atlas selection process with respect to the normalized correlation coefficient metric. The similarities obtained between the target image and the warped images are ordered from the highest to the lowest, to perform the majority voting process with the best mask and add one by one the next best ones until all the images in the atlas are used.

Figure 1. Code example for de-identifying DICOM files.

Figure 2. Example of a YAML de-identification profile.

Visualization of one shot of the spin-echo sequence using MRSequoia’s built-in visualizer. Zooming and panning can be achieved with the two sliders or the buttons adjacent to it. In the table underneath, information about the sequence atoms at the position of the vertical blue line at the center of the plot is displayed.

Code to import the sequence timing from the Siemens IDEA simulator output and to perform a check that TR is consistent across all shots. The selector extracts all RF pulses (represented by class TXAtom) with a flip angle of 90°. The check then consists in verifying that the temporal spacing between subsequent pulses is constant. The output of the tests indicates that 129 such TX pulses were found (one dummy + 128 phase-encode shots), and that all of them are spaced 5 seconds apart, thus indicating success of the test.

Figure 1: (a) Organization of the three components of the EPG algorithm, (b) 100 signals simulated using various values of T1 and T2 for a multi-echo spin-echo sequence with 60-degree refocusing flip angles.

Figure 5: The simulator is combined with a fully connected neural network for estimating T1 and T2, where the flip angle train is optimized via auto-differentiation to minimize fitting error.

Figure 4: a): Electrode-specific corrective factors for the total field magnitude at the WM surface, mapped to the electrode regions on the skin surface. b): Error between the scaled electric field magnitude for the test case and the electric field magnitude for the base case at the white matter surface, mapped to the electrode regions on the skin surface.

Figure 5: Magnitude of the total E-field induced by electrode C4 just outside the WM surface. The E-field magnitude has been clipped to 1.4 V/m in all cases. a): E-field is presented for the simplified case in which both the diploë and dura are treated as cortical bone. b): E-field scaled by the corrective factor is presented for the simplified case. c): E-field is presented for the base case where the diploë and dura are assigned their own electrical properties.

The three types of users supported by DGF. User A requests simulations from the web GUI and runs them on a dedicated Cloud MR AWS account. User B runs the simulations from the web GUI using her/his own cloud computing account. User C runs the simulations on a local computer. Users B and C can synchronize the results on Cloud MR using restful API’s. All types of users can display results on the web GUI or download them in standard formats (e.g., mat or json).

The Set Up tab of DGF’s web GUI. The web GUI is implemented in html and uses several javascripts libraries (angularjs, jquery, bootstrap, plotly.js, math.js and Three.js). The Set Up tab is divided into panels that enable customizing different aspects of the simulations. This figure shows the Object panel, from which users can define the properties of the various object layers (three in this example) and specify the size and position of loop coils.

Figure 1: A) Shows a flowchart of the process for using GrOpt to build waveforms to enter into Pulseq. B) Shows how the system and hardware constraints are initially shared to initialize the GrOpt optimization and Pulseq limits in the code example.

Figure 3: Summary of designing PC-MRI sequence with GrOpt and Pulseq. A) Gives a code snippet of relevant portions of the sequence design code. Some initial gradient and RF creation omitted for space. B) Shows the waveform designed by GrOpt. C) Shows the PNS response of the B), where the PNS does not exceed the 0.8 limit imposed. The optimal design slews quickly before the PNS limit, then the slew rate derates accordingly. D) Shows two TRs of gradient waveforms as output by Pulseq, with the GrOpt bipolars played out in the z direction after slice select.

Figure 1. Image of patient and healthy control (HC) with simulated lesions of patient. Left panel, from left to right: native 3DT1 patient image; and native image with the transformed lesion mask (red). Right panel, from left to right: native 3DT1 HC image; HC image with simulated lesions; and HC image with the simulated lesion mask (red). Note that the lesion mask of the patient was manually outlined on a FLAIR image and transformed to the T1 image with nearest neighbor interpolation. GM folds (as visible in top left corner of HC axial slice) are not segmented as lesions.

Figure 3. Intensity plot of the intersection of a patient lesion (blue) and a simulated lesion (red) with surrounding tissue. Note the intensity axes are different for the patient lesion and simulated lesion.

Fig.3. The four-dimensional numerical phantom for the Bloch simulations of a water-fat system. The 3D water phantom was used for the Bloch simulation in the uniform magnetic field corresponding to -114 Hz resonance frequency. The 3D fat phantom was used for the Bloch simulation in the uniform magnetic field corresponding to +114 Hz resonance frequency. The simulated images were reconstructed after adding the two sets of MR signal obtained for the water and fat phantoms.

Fig.4. (a) The experimentally acquired central cross section of the block bacon for the in-phase and out-of-phase conditions. (b) The simulated central cross section of the block bacon for the in-phase and out-of-phase conditions.

Figure 5. Reconstruction of data acquired imaging an ISMRM phantom. A single-slice 2D Gradient Recalled Echo sequence was designed using PyPulseq on Colab. The sequence was then exported as a .seq file and executed on a Siemens Prisma Fit 3T scanner. The acquired data was reconstructed using a 2D Fourier transform.

Figure 2. Screenshot of installing the prerequisite PyPulseq library and importing the required packages. The ‘about’ section of the Colab notebook links to a reputable resource where users can learn more about the sequence. In the ‘install’ section, PyPulseq is installed from Github via a single command. The ‘import’ section details the specific components that are leveraged in constructing the 2D Gradient Recalled Echo sequence.

Figure 4: Example RIESLING reconstruction of 3D radial IR-ZTE dataset. Comparison of the three reconstruction methods: root-sum-of-squares (RSS), CG-SENSE, and Total Generalised Variation (TGV) at two different acceleration levels relative to Nyquist sampling criterion.

Figure 1: Overview of the riesling toolbox including the riesling .h5 file format and the available tools.

Encoding and tabular files for a Stejskal-Tanner sequence. A series of volumes (upper left) are labelled with index t (table, lower left). These are defined with a prototypical sequence (right), and scaled and rotated by indices s and (x,y,z) respectively (bottom left, bottom middle).

Structure of files within an aDWI-BIDS folder (additional files bounded in blue). The encoding file describes a prototypical sequence, which is allocated to acquisitions given in the tabular file. The tabular file in turn may describe perturbations to the prototypical sequence on a parameter by parameter basis. Perturbed sequences describe slices or volumes within the NIfTI (figure 2).

Fig. 2 - Kurtosis metrics for a representative axial slice of the CFIN data: (A) mean diffusion; (B) radial diffusivity; (C) axial diffusivity; (D) fractional anisotropy; (E) mean kurtosis; (F) radial kurtosis; (G) axial kurtosis; (H) Mean signal DKI index. The mean kurtosis tensor is relatively robust to noise-related artifacts that appear in standard MK (red arrows).

Fig.4 - Metrics from the conversion of DKI to the spherical mean technique (SMT) two compartmental model^9,11: A) axonal water fraction (AWF); B) intrinsic diffusivity (ID); and C) microscopic fractional anisotropy (μFA).

Figure 4. Prototype MRSDB user interface. The interface allows for uploading, filtering, visualizing, and downloading of specific data points for further analysis or evaluation of normative ranges per population.

Figure 2. Distribution of demographic, system and site information of the MRSDB reference library. This includes (a) 12 standard brain locations commonly measured with MRS, (b) a broad range of echo times (TE), strongly focused on short TE, (c) 70% male scans and 30% female scans, (d) six different scanner types (Siemens), and (e) eight different software versions.

Figure 2: Example B₀, B₁, T₁, T₂, R₂^* and ADC maps generated from the harmonised UKRIN-MAPS protocol using UKAT.

Figure 1: An overview of the modules in the UKAT toolbox showing development progress and future plans.

Flow of pipeline from raw DICOMs to end-user entry into a Research Electronic Data Capture (REDCap) study management database. XNAT: Extensible Neuroimaging Archive Toolkit, HPC: High Performance Computing system, Slurm: Slurm Workload Manager, HeuDiConv: Heuristic-centric DICOM Converter, MRIQC: Magnetic Resonance Imaging Quality Control, fMRIPrep: Function Magnetic Resonance Imaging Preprocessing, xcpEngine: the XCP Engine, BCT: Brain Connectivity Toolbox.

Comparison of global efficiency as calculated by the Brain Connectivity Toolbox MATLAB functions for two common structural atlases (Automated Anatomical Labeling 116, Desikan-Killiany) and the Power 264 functional atlas after three methods of denoising in xcpEngine (36 motion parameters, 36 motion parameters and Power scrub method, and ICA-AROMA).

Figure 2. (a) Results of rigid registration displayed at the slice near the iso-center (red-MR and blue-CT), (b) 3D rendering of the representations of the gridpoints' coordinate pairs (MR-magenta and CT-green), (c) coordinate pairs superimposed on MR image, (d) coordinate pairs superimposed on CT image, and (e) distortion plot against the distance from the iso-center.

Figure 1. (a) User interface of the developed distortion measurement prototype with qualitative visualization; (b) user interface with quantitative evaluation; (c) backend pipeline of the developed distortion measurement prototype.

Figure 1: Overview over the steps performed by our automatic reproducibility test. An example of a config file can be seen in Fig 2.

Figure 2: Configuration file for reproducing the demo data included with [5], showing the NRMSE-Tolerance, the reference version of BART and the test branch, as well as the scripts (in this case just "all.sh") and the outputs to be tested (recon-turnbased and reco-goldenangle). While not a full reproduction of all results of that publication, this configuration file can serve as an illustrative example.

Figure 1: Brain MRS - Adjustments. The simulated brain spectrum for 3 T (A). The spectrum can be adjusted for echo time from 0 to 300 ms (B), line broadening (C) and noise (D). Screenshots from the iOS simulator.

Figure 2: Brain MRS - Analysis. User can control echo time, line broadening and noise levels in order to create a realistic spectrum (echo time = 20 ms, line broadening = 5 Hz, noise = 80) (A). The spectrum can be analyzed by visualizing a metabolite, e.g. background signal of macromolecules (B), or group of metabolites, e.g. N-acetyl aspartate, myo-inositol, glutamate and glutamine (C). Screenshots from the iOS simulator.

Fig.2: MeVisLab graphical user interface showing basic modular concept: 1),2) a single MRI BART toolbox command (no inputs/outputs) for simulation of Shepp-Logan phantom data with multiple different coil sensitivities with graphical control unit, 3),4) readout module for output of toolbox data with control unit, 5) MeVisLab visualization modules for interactive data visualization, 6) MeVisLab interactive viewer for medical imaging data

Fig.1: Top: Schematic of the set-up: Via the MeVisLab UI a user can control remote MRI reconstruction and post-process any result with tools of the MeVisLab toolbox. In the first step MRI data must be acquired and the raw data must be transferred from the MRI scanner to the computing device. After composing a remote reconstruction pipeline in the graphical user interface, remote computation is launched and results can be analyzed in each step of the pipeline. We have implemented a remote service to control MRI reconstruction frameworks via docker containers.

Figure 1: The simulation pipeline enabled by bridging Pulseq to JEMRIS. Sequence, phantom, and B1 map files are converted into JEMRIS format, simulation is performed on the JEMRIS command line, and the resulting data is reconstructed automatically to yield images.

Figure 4: Simulating and acquiring 64 x 64 phantom images from the same sequence file (FOV = 250 mm, slice thickness = 5 mm, TR = 4500 ms, TE = 50 ms). Simulation at FA = 70 degrees was performed while images were acquired with FA = 45, 70, and 90 degrees. At FA = 70 degrees, PSNR = 19.12 dB and SSIM = 0.7531 between the simulated and the acquired images.

Figure 2. B0 field maps in a uniform cylindrical phantom (Site 1). Left image in each sub-panel: Acquired field map after running the scanner’s built-in linear shim routine. Right image in each sub-panel: Acquired field map after performing 2nd-order shimming using the proposed toolbox. 9 slices are shown.

Figure 3. B0 field maps in an anthropomorphic head phantom (Site 1). Left image in each sub-panel: Acquired B0 field map after running the scanner’s built-in global linear shim routine. Right image in each sub-panel: Acquired B0 map after performing localized 2nd-order shimming using the proposed toolbox over a frontal 3D region (in vicinity of purple arrow). 8 slices are shown.