Figure 5. Results of in-vivo experiment. (a) The estimated motion parameters using the 4D navigator compared with the reference estimation using high-resolution volumes. (b) Comparison of the estimated B₀ changes relative to the first TR at two different head positions using the proposed navigator and reference acquisition. (c) Reconstructed quantitative parameters using 3D-EPTI data without motion (first row), with motion but without correction (middle row), and with motion correction (bottom row).

Figure 1. (a) Illustration of the IR-GE 3D-EPTI acquisition with 4D navigator. In each TR, a 4D block is acquired at each TI. 20 TIs are acquired for IR-GE with a golden-angle radial-block sampling, followed by additional 4 TIs for navigation that utilizes the same sampling in every TR. The optimized spatiotemporal encoding used for the block-wise 4D acquisition is shown in (b). 3D motion parameters and B₀-inhomogeneity changes across the whole brain can be estimated from the reconstructed multi-echo navigators.

Figure 4: Comparison of image quality without motion correction, with MERLIN motion correction (MOCO), and without motion. A clear improvement in image quality is observed with MOCO.

Figure 3: Animated figure showing the estimated translations (Δ_x,Δ_y,Δ_z) and rotations (α_x, α_y, α_z), which show a pattern similar to the motion paradigm described in Figure 2A. Gray section in the beginning indicate dummy and low-resolution spokes, for filling the dead-time gap.

Figure 4: Magnitude (A) and phase (B) sagittal slices of the cerebellum. C-D, zoomed views. Note the white stripes in D (examples with black arrows), next to the WM, as reported before in². These stripes were consistent across successive slices (E,G,I: magnitude, F,H,J: phase) and are likely to be the granular ceberellar layer. Note they fell within the cerebellar gray matter, as seen in the magnitude image (blue bar in I,J).

Figure 2: Sagittal slices of the cerebellum (magnitude image; 0.2x0.2mm² in-plane resolution). Each row is a participant; left column: PMC off, right column: PMC on. The white boxes show zoomed views of the cerebellum. AES (average edge strength) is a measure of sharpness; FD (framewise-displacement) is a measure of shot-to-shot motion as determined posthoc by the fat-navigators. Note the marked improvements in image quality when using PMC.

Figure 2: Hybrid k-space (left) containging pilot-tone signal and amplitude (middle) and phase (right) traces with motion encoded in each coil. Each color line is a channel. The data shown is from the DISORDER set.

Figure 5: DISORDER reconstructed images. Top row: Un-corrected; Middle row: Pilot-tone motion corrected; Bottom row: Pilot-tone motion corrected + corrected for pose-dependent fields.

Figure 1 (animated): High temporal-framerate PET reconstruction (Δt=1s) in the transient phase ~8mins after the injection of the Ammonia PET tracer. Corresponding respiratory waveforms (bottom) obtained using either Principal Component Analysis (PCA, black) or a pencil beam navigator over the lung-liver interface (red). This patient demonstrates a deep, regular diaphragmatic breathing pattern.

Figure 2 (animated): ZTE lung images corresponding to the Ammonia PET tracer patient shown in Figure 1. Because of the deep diaphragmatic breathing the uncorrected (averaged, left) images show strong motion blurring (especially at the lung-liver interface). Soft-gated respiratory binning (7 phases, 2^nd column) resolves the diaphragmatic breathing cycle into 7 phases. Most of the data are acquired in the end-expiratory phase (3^rd column) also providing the sharpest image. Its Maximum Intensity Project (MIP, right) depicts the vascular anatomy and lung lesions in fine detail.

Figure 1: Alternating optimization is computationally demanding as repeated updates of the motion vector θ and image estimate x are needed. SAMER speeds up the optimization by utilizing a scout scan as an image prior x_sc. This avoids the need for repeated updates of x during the motion estimation. In this work, SAMER is extended to 2D TSE imaging and motion parameters are estimated from a low-resolution 3D SPACE scout. This is feasible as each shot of the imaging scan (green circles) has common frequency overlap with the low-resolution scout (dashed orange box).

Figure 4: In Acq. 1 (mostly in-plane rotation), SAMER mitigated most motion artifacts, which allowed fine anatomical structures, such as a blood vessel to be recovered (yellow arrow in zoom-in). The image quality improvement is also reflected by a decrease in data-consistency error (DC). In Acq 2. (through-plane rotation), SAMER also yielded robust artifact reduction, however, the zoom-in shows a small loss of spatial resolution (red arrow).

Figure 2. Automated motion estimation and reconstruction pipeline. The network consists of $$$N$$$ iterations between motion estimation blocks and a registration-based 3D reconstruction block. (a) Motion estimation block. (b) The proposed deep recursive framework, including motion estimation blocks from three orthogonal orientations and a rigid slice-to-volume registration-based 3D reconstruction transformer.

Figure 3. The performance of different motion estimation algorithms on the simulated motion-corrupted data. (a) Mean absolute error (MAE) and root mean square error (RMSE) metrics of rotation angle. (b) One typical case (31-week GA fetus in the testing set) of the motion correction using deep recursive framework (DeepRF). DeepPE: Deep Pose Estimation, DeepPMT: Deep Predictive Motion Tracking.

Fig. 1: Proposed LAPNet to perform non-rigid registration in k-space. Moving ν_m and reference ν_r k-spaces are tapered to a smaller support W. The bundle of k-space patches is processed in a succession of convolutional filters (kernel sizes and channels are stated) to estimate the in-plane flows u₁,u₂ at the central voxel location determined by the tapering T/window W for size 33x33x33. Overall 3D deformation field u is obtained from a sliding window over all voxels in orthogonal directions.

Fig. 2: Respiratory non-rigid motion estimation in a patient with a liver metastasis in segment VIII. Motion displacement is estimated by the proposed LAPNet in k-space in comparison to image-based non-rigid registration by FlowNet-S (neural network) and NiftyReg (cubic B-Spline). Estimated flow displacement are depicted in coronal and sagittal orientation. Undersampling was performed prospectively with a 3D Cartesian random undersampling for 8x and 30x acceleration.

Figure 4. Reconstructed slices near LAD and RCA. The proposed method exhibited improved sharpness not only near the coronary arteries (orange arrows) but also in non-cardiac regions (green arrows) compared to 3D translational motion-corrected reconstruction.

Figure 1. Overview. (a) For each heartbeat, we collect segmented cones interleaves that sparsely cover the entire k-space along with another set of cones interleaves for the 3D iNAVs. (b) 3D translational motion is estimated using the iNAVs. Segmented cones data from individual heartbeats are binned to reconstruct image-based self navigators. (c) Nonrigid motion is estimated using an existing image registration method. (d) The nonrigid motion-corrected reconstruction is achieved by solving an optimization problem using image-space gridding operators.

Figure 3. Temporal dynamics of all 17 DCE contrasts reconstructed using FF MoCo P2P MVF on the patient shown in Figure 1-2. Each contrast was reconstructed using 20 seconds of non-overlapping data.

Figure 2. Example contrast 1 pre-injection 3D MoCo images reconstructed using Spatial MoCo and FF MoCo with MCNUFFT, CS and P2P MVF.

Figure 1. Overview of DeepResp. (a) Overview of DeepResp. DeepResp is designed to extract the phase error from a GRE image. (b) Architecture of a two-stage neural network in DeepResp. The first stage extracts the differential values of the phase errors. The second stage accumulates the differential values, generating the phase errors.

Figure 2. In-vivo results of DeepResp in deep breathing. (a) Two slices (first and third rows) and their zoomed-in images (second and fourth rows) are shown. Artifacts (red and yellow arrows) are clearly reduced in DeepResp- and navigator-corrected images. The phase errors show very high correlations between the results of DeepResp (red line) and the navigator (black line). (b) Quantitative metrics of NRMSE report improvements in all subjects (red: DeepResp-corrected; blue: uncorrected images)

FIGURE 1 (a)The overall architecture of the proposed cGAN-based method.(b) The discriminator framework, consists of 5 cascaded convolution layers.(c) The generator framework, it contains 5 encode and 5 decode block, and 7 cascaded Resblock.(d)The entropy curve, shows that the motion amplitude increase, the entropy increase.

FIGURE 3 Mild and strong motion correction results of various methods on multi-shot FSE sequence. Columns of the first is the mild motion result and the third is strong. The columns from left to right show reference images, motion-affected images, motion corrected images using TV Denoiser, DnCNN, UNET, and our method, respectively. In the second and fourth row, the absolute error maps corresponding to the first and third row are presented. In each motion corrected image, the correction quantitatively (PSNR, SSIM, MSE) can be seen in comparison to the motion-free reference.

Rotation plus translation results. The first row of each subfigure contains clean image, corrupted image, motion correction result of L1, motion correction result of L1+TV, and motion correction result of L1+TV ft, respectively. The second row of each subfigure shows motion trajectory, residual image between corrupted image and clean image, residual image between motion correction result of L1 and clean image, residual image between motion correction result of L1+TV and clean image, and residual image between motion correction result of L1+TV ft and clean image, respectively.

(a) SSIM (mean ± std) of the motion-corrupted images, L1, L1 + TV, and L1+TV-ft, respectively. (b) The proposed method successfully reduced the effect of motion. SSIM against the reference image is plotted as a measure of image quality (0 is the lowest, 1 is the highest). The quality of MC-Net predication is similar for input images with only rotational motion and those with both rotational and translational motion.

Reconstructed images with CG-SENSE and the proposed method, and ground truth images. Compared to CG-SENSE, noises are reduced by the proposed method. (A) CG-SENSE, (B) the proposed method, (C) ground truth. (D)-(F) Magnified images of the solid boxes in (A)-(C), respectively.

Schemas of parallel imaging (PI) reconstruction of blade images for our accelerated PROPELLER. The neural network achieves a higher PI factor than the standard SENSE reconstruction, enabling faster acquisition.

FIG. 1: Schematic description of DLnav processing.

FIG. 3: Respiratory waveforms and signal-to-noise ratio of respiratory-like frequencies (SNR_R) calculated with the conventional (green) and DLnav methods (yellow). a, b: Respiratory waveforms overlapped on the navigator signals with the tracker at the normal position (a) and the tracker on the heart (b). c: SNRR with the tracker of the normal position (Normal) and the tracker positioned on the heart (Heart). DLnav showed a good synchronization with the actual respiratory motion even at the bad tracker position on the heart.

Figure 3. Mean absolute error (MAE, mean and standard deviation) and structural similarity index (SSIM, mean and standard deviation) comparing the ground truth results to the uncorrected images (black), as well as images corrected with the combined (yellow), single-channel (blue), and multichannel (red) models. All differences are significant (p < 0.05).

Figure 4. Example images for each contrast; T2-weighted (top), T1-weighted (middle), and FLAIR (bottom). On the left are the ground truth images, and on the right those containing simulated motion artefacts. The centre three images are those corrected with the combined, single-channel, and multichannel models (from left to right).

Fig. 5: Reconstruction of the motion corrupted k-space by removing or replacing the highest ranked ET by the NN (RMSE: red; SSIM yellow). The curve in the plot below shows the ground truth motion; the numbers represent the ranking of the motion severity obtained by the NN. For simplicity only rotation is simulated in this example.

Fig. 3: Structure of the Neural network with respective output sizes (left) and the visualization of the MS (right).

Figure 1. Motion simulation strategy for spiral sampling (A), and network architecture adopted in this study (B).

Figure 2. Representative motion compensation results on simulated data, from subjects 1 (A), 2 (B), and 3 (C).

Figure 1. Proposed rigid motion correction process.

Figure 4. The resultant mGRE images corrected from real motion-corrupted images.

Figure 1: a) An overview of the proposed method. The input images can contain motion-free images and motion-corrupted images. The output images are the motion-corrected images if the corresponding input images are motion-corrupted, whereas the outputs are almost the same as the input images if the corresponding input images are motion-free. b) The detailed framework of the proposed network with the network hyperparameters.

Figure 3: Example images of the baseline experiment for a) the synthesized test data and b) the real motion data in T1w, T2w, and FLAIR correction cases. SSIM and NRMSE scores are written on the motion-corrected images. For the real motion test, the red and blue boxes highlight the performance of the proposed method.

Figure 5. In-vivo evaluation in fetuses at 32 and 31 weeks' gestation. The fetal brain is arbitrarily oriented relative to the planes of the MRI scanner. The network detects the brain, left (red) and right (green) eyes from the full-uterus scout, oriented along the device axes (sagittal to the mother). From these landmarks, we derive the head pose. The target sequence automatically acquires standard sagittal, coronal and axial views of the brain. We repeat the scout before each anatomical acquisition. The second fetus required several attempts due to substantial subject motion.

Figure 2. Automatic field-of-view prescription. First, the operator acquires a rapid full-uterus scout. Second, an external laptop receives the scout via TCP and hosts the network that detects the fetal brain and eyes. Third, we construct an anatomical basis from these landmarks. Finally, the target sequence receives the anatomical frame and acquires sagittal, coronal or axial slices of the brain according to the operator’s selection. All communications are automatic, and the procedure can be repeated as needed to acquire different views or to respond to fetal head-pose changes.

Object tracking on (a) a coronal and (b) a parasagittal image series of the abdomen and thoracal borders (3T, TrueFISP, FOV 384x384 mm², 256x256 matrix, TR/TE 360/1.46) with bounding boxes independently tracking different landmarks on the images (diaphragm, liver, spleen, kidney). (c) Object tracking performed on a CINE acquisition of the heart. The target is missed during end-systole (red boxes) when the image features change, but reestablishes the position back in diastole.

Tracking of the heart on a series of fast low-resolution image navigators (sagittal, 256 repetitions, every 8th displayed) as appropriate for prospective motion correction of cardiac MR spectroscopy. The bounding box follows the heart well during several respiratory cycles, with heavy breathing starting from repetition 100. The detected in-plane motion, as indicated by the series of arrows on top, is exaggerated by a factor of 3 for visualization purposes.

Figure 1: Geometry of the virtual phantom: The volume “Field sensor volume” was voxelized at $$$0.8\,mm$$$ isotropic resolution. Voxel size outside this volume was increased automatically where possible. The generator loop is labeled as a white circle. The blue plane represents the analysis plane $$$\Sigma_{ant}$$$. Red and green circles show the location of virtual coils $$$L_{R1}$$$ and $$$L_{R2}$$$ used in the further analysis. Boundary volume was set automatically to $$$(1366\times1217\times1250)\,mm^3$$$ using uniaxially perfectly matched layer (UPML) condition.

Figure 2: Cardiac volume ground truth (a) and simulated received time-signals $$$u[t_n]$$$ in the virtual coils $$$L_{R1}$$$ (b) and $$$L_{R2}$$$ (c) plotted as modulation depth, i.e. change relative to the mean signal in percent. The approximate position of ECG R-peak is shown in orange in (a). Black arrows in (a) show periods of contraction (systole) and expansion (diastole) in the cardiac volume curve.

The figure shows the simulated experimental set-up. Two coils are fixed onto the end-pieces of a pair of glasses. Each solenoidal coil was composed of 100 turns of 0.25mm- diameter wire arranged in 10 layers radially, each formed from 10 turns. The left and right coils are oriented along the y- and z-axes, respectively. The z-component of the magnetic field from the coils is monitored using 16 field probes spanning 21 cm axially and 22 cm azimuthally.

Coil positions (top) and z-component of the magnetic field⁴ on a cylindrical surface (bottom, B_C ) (0.3A coil current). Probe positions are highlighted using circles (top) or line crossings (bottom). The field at the initial head-position is shown (left), along with the field changes produced by head pose changes during head nodding (middle) and head shaking (left) (motion parameters shown in Figure 3. Complex patterns of field change give high sensitivity to differences in motion.

Figure 4. a) Correlation between the calculated motion score and the user score for the development set. b) Correlation between the calculated motion score and the user score for the test set, with an R² value of 0.91

Figure 5. Sample images taken from each of the nine series in the test set. U = The user assigned score. A = The algorithm assigned score

Figure 3 Three slices from axial, coronal and sagittal planes to demonstrate the motion artifacts and the effect of motion correction using radNAV. After correction, the least square error reduced from 16.96% to 12.44% and the structural similarity index (SSIM) increased from 83.75% to 89.28. Red arrows indicate the motion artifacts before/after the correction.

Figure 1 A: Sequence diagram for turbo-FLASH with radNAV; B and C: k-space trajectories for one navigator frame. B: sequentially ordered spokes with 15 (black dots) and 30 (red circles) evenly distributed in 3D space as temporal period. C: 15 (black dots) and 30 (red circles) 3D golden angle spokes. Different spoke number in each navigator frame for sequential order (B) completely alters k-space trajectory, while 3D golden angle spokes (C) overlaps given different spoke number, suggesting a sliding window method is applicable for increasing temporal resolution of the navigator frame.

Figure 1: (a) Joint correction alternates between image and motion estimation for all shots. (b) Incremental correction estimates the initial image using shots with similar motion. Then motion for each shot is sequentially estimated followed by image update. (c) Initial image used in incremental correction has much lower artifacts than the initial image used in joint correction. The final estimate in incremental correction provides the motion corrected image.

Figure 2: Results from the simulation study. (a) iMoCo was able to recover the image without significant residual artifacts, (b) Images from joint correction were improved as the number of iterations were increased. However, residual artifacts were seen (arrows) even after 80 iterations because the solver had not converged as indicated by the data consistency error plot in (c).

Fig. 3. SMS-EPI reconstruction of SSG and SENSE with original coil sensitivity maps (oCSM) and updated coil sensitivity maps (uCSM) at maximum motion velocity (a), and corresponding motion parameters for single TR period (b). Images reconstructed without uCSM exhibit aliasing artifacts, as indicated in zoomed views (red boxes). Conversely, reconstructions with uCSM show clearer inner structure and boundaries as well as much less aliasing artifacts.

Fig. 4. SSG and SENSE reconstructions of stationary phantom (x/y/z-translation -2.4/6.4/21.5mm; x/y/z-rotation 10.5°/5.2°/-0.5°). Residual aliasing artifacts appear in most slices of the SSG with oCSM (white arrows), whereas the SSG with uCSM show reduced artifacts and clearer inner structure and boundaries. SENSE images with oCSM show aliasing artifacts (red boxes), while uCSM improves delineation of structures and eliminates slice-aliasing artifacts.

Figure 1: Self-navigating 3D-EPI sequence

Figure 2: Volumetric images acquired (a) before motion occurred (reference volume), (b) when motion occurred, and (c) the next volume after motion occurred.

Figure 3. Representative motion tracking and quantitative maps of a healthy volunteer with intentional in-plane (“side-to-side”) head motions. The head motion was rigidly tracked in three translational and three rotational directions. (a) Motion tracking time curve of translations and rotations in the x-y-z coordinate system. (b) T1 and T2 maps acquired with the proposed method and those without motion correction are shown.

Figure 5. Effect of motion correction on regional quantitative values. (a) Bland-Altman plots representing the bias of scans with motion with and without motion correction compared with scans without motion as references. Data points with motion correction are closer to zero than those without, indicating smaller bias achieved by motion correction. (b) Coefficients of variation (CV) represent the repeatability of the scans. In both T1 and T2, within-subject CVs were smaller, indicating higher repeatability, in motion-corrected scans than those without motion correction.

Figure 4: In-vivo examples from 4-interleave 3-SMS full-volume DWI reconstructions. a: SMS-IRIS and MoSaIC examples. b: Shot-wise shift and rotation parameters for MoSaIC (see color code). In the static case (blue), the parameters are almost constant and the images appear similar. The SMS-IRIS images of the next two examples (orange and green) are visibly blurred by in-plane motion, which is improved by MoSaIC. The last two SMS-IRIS images (red and purple) are mainly affected by through-plane motion, which is detected as x-rotations and y-shifts and improved by MoSaIC.

Figure 1: MoSaIC scheme for navigated DWI reconstruction per diffusion direction including 3D rigid motion correction. SMS navigators are unfolded using 2D-SENSE. The shots from the first diffusion-weighted TR are assumed motion-free and stacked to the reference volume. The navigator SMS groups are registered by SMS2Vol registration, the shot diffusion phases are extracted and rejection criteria are calculated. All shot parameters are included into a motion-corrected full-volume reconstruction combining the high-resolution image data.

Figure 1: Example images and motion estimations from a 7.6 year old typically developing female demonstrating MPnRAGE motion correction for moderate to severe motions. Without motion correction, the T1-weighted and quantitative R1 images have blurred tissue boundaries.

Figure 3. The regional surface maps for the coefficients of variations x 100 without (top) and with (bottom) motion correction.

Figure 5. Axial maximum intensity projections (MIPs). The MIPs are of very high quality with detailed depiction of distal vessels and excellent background suppression in the cases of no intentional motion (A and B). The trained motion causes clear misregistration between vessels in the brain (C, red arrow). Further, many distal vessels are lost on the motion corrupted image (C, yellow arrow), and there are artifactual stenoses (C, green arrow). With prospective motion correction, the misregistration and the distal vessels are mostly recovered, and pseudostenosis is corrected.

Figure 4. Sagittal (A, B, C, D) and coronal (E,F,G,H) maximum intensity projections (MIPs). Without intentional motion (A, B, E, F), activation of PMC does not substantially alter image quality. With trained motion, some vessels in the middle slab are obscured (yellow arrows) (C,G) without PMC. There are artifactual stenoses between slabs (green arrows). There is also misalignment between slabs (G, red arrow) and discontinuous major vessels, e.g. the middle cerebral arteries (G). In comparison, the obscured vessels are recovered when PMC was on and misalignment alleviated (D, H).

Sagittal field maps generated from the multi-echo FLASH and dual-echo vNav acquisitions shown in Hz. The FLASH field maps were downsampled by a factor of 4 to match the resolution of the vNavs for easier comparison. Note that the field map values have not been unwrapped, so may not be entirely comparable in regions of very high susceptibility due to differences in echo timing between the FLASH and vNav scans.

Sagittal views of magnitude images from the FLASH and vNav scans, for each of the first and second runs. For the first run, the subject's head was held in a fixed neutral position, and for the second run, the subject's head was held in a fixed upward nodding position.

Figure 5. Each coloured region in the plot bounds the rotational and translational motion parameters range for each evaluation category after FatNav (left) and without motion correction (right), in case of smooth (top row) and rough motion (bottom row). FatNavs can correct very well for an RMS value, averaged along the three axes, of ~3.7°/3mm and 2°/1.6mm for smooth and rough motion respectively (category 4 boundary); without motion correction, image quality drops much more quickly (~1.2°/1mm).

Figure 1. Comparison between FatNav volume before and after GRAPPA ‘re-reconstruction’, for four different amounts of motion (tables on the left), to simulate effect of mismatched ACS data: parallel imaging artifacts increase with the amount of motion.

Figure 1. (a) The setup for phantom test. (b) Three markers were stuck on a wooden rod, and then attached on the base plate of the MR motion phantom, which can produce smooth linear motion. (c) Left: wireless tracking marker proposed in reference ^[2]; Right: marker used in our method. (d-f) Demonstration of our homemade head strap. Markers were placed in plastic holders and then attached on the headbands.

Figure 4. Tracking results of in-vivo experiments. (a-c) Measured tracking traces along three directions (LR: left-right, SI: superior-inferior, AP: anterior-posterior) when head shaking was performed. (d-g) Measured positions in 3D space for three markers at two selected time points (red and green markers in Fig.4a).

Figure 1. An example of a gradient echo sequence timing diagram. Red dotted lines mark possible block boundaries. RF pulses and ADC events are not allowed to cross block boundaries; however, gradients are not forced to 0 at the boundaries (see blue circles), allowing for efficient gradient wave forms as needed for the fast sequences. Block arrows on the top mark possible position update points; filled ones correspond to the currently implemented option.

Figure 2. In vivo images acquired with a 3D gradient echo sequence programmed in Pulseq in presence of head rotations: (left) without prospective motion correction and (right) with prospective motion correction. See Fig. 3 for corresponding motion traces.

Experimental set-up. The NMR field probes were placed between the transmit and the receiver Rf coils. The optical camera (MPT, Kineticor) is fixed to the inside of the magnet bore. The positions of the probes are shown in (b).

The plots show examples of simultaneous measurements of head motion parameters (top row) and magnetic field (bottom row). The subject performed various head movements (rest, head shaking, head nodding, feet wiggling). Measurements were acquired with and without simultaneous EPI scanning.

Fig. 1 The setup of SLOMO system for abdominal imaging. SLOMO system consist of frame holder (a.) and optical module (b). Optical module is a camera-laser system for depth imaging

Fig. 4 Reconstruction results of motion corrupted and corrected images. Liver images were reconstructed into 4 phases according to the respiratory curves detected by bellow and SLOMO system.

Fig.4: Resulting in vivo fat-water and fat fraction ($$$\text{FF}$$$) images of a coronal slice are shown for non-respiratory, non-cardiac resolved (NRR) and respiratory motion-corrected, cardiac binned (MOCO) reconstruction. The latter is shown in the end-diastolic phase. The red arrows indicating a blood flow artifact visible in NRR and reduced in MOCO. Line plot position is indicated by the dotted lines (I) and (II) with increased $$$\text{FF}$$$ for MOCO up to $$$24\%$$$ and reduced blood flow artifacts by the factor of 2 compared with NRR.

Fig.5: Fat-water and fat fraction ($$$\text{FF}$$$) images for 3 orthogonal views are shown for all 10 volunteers.

Figure.2 A comparison MOLLI T1mapping with/without MoSe-CINE approach and FEIR. Both conventional MoSE-CINE visual approach with FEIR clearly improved the accuracy on T1 confidence map and the combination of MoSE-CINE visual approach with FEIR showed the best image quality.

Figure.1 MoSe-CINE images allows direct visualization of motion-independent cardiac phase timing. It shows depicted signal decrease due to cardiac motion and it pointed out when is the best timing to trigger for both systolic and diastolic timings.

Figure 3. Comparison of computational times in minutes per patient (1-10) for GFMR (red) and MDR (blue) methods on T1 (top), DTI (middle) and DCE (bottom).

Figure 2. Distribution of pixel-based median metrics (one per row) for all 3 contrast mechanisms (one per column) and for each individual subject (horizontal axis). Plots show median +/- standard deviation for uncorrected data (green), GFMR (red) and MDR (blue).

Figure 1: Typical motion in T2-weighted imaging with averaging
(a-c) single average with degraded image quality and displacement shift
(d) combination of all averages without MOCO
(e) combination of all averages with MOCO and improved image quality

Figure 2: Improved image quality of MOCO
(a) conventional reconstruction (Likert-score 3)
(b) MOCO (Likert-score 1)

Figure 1. Flow diagram of the iMoCo process. Redundant sampling of the b50 volumes results in well-filled motion states, allowing for accurate non-rigid registration. Motion vectors produced by the non-rigid registration are then applied to the remaining b-value volumes, which do not require redundant sampling.

Figure 3. Motion Phantom (MP) and Healthy Volunteer ADC Maps (Voxel size 1.5x1.5x5.0 mm3, PAT 2, Matrix Size 128x104x35, FOV 380 mm, TR 6.1 s, TE 56 ms, b50-16 avgs, b800-16 avgs, BW 2300 Hz/Px). Motion Phantom diffusion values are most comparable between the gold-standard Gated and iMoCo maps.

Figure 2: Comparison of ¹¹C-UCBJ PET images reconstructed with conventional BSREM (top row), MR guided BSREM using the anatomical priors without motion correction (middle row) and MR guided BSREM with motion correction (bottom row). MR guided BSREM shows better SNR and higher image resolution compared to the BSREM method; however, as shown by red arrows, incorporating motion correction into MR guided BSREM (bottom row), results in a sharper image with crisper edges.

Figure 4: Comparison of ¹⁸F-PI2620 PET images reconstructed with conventional BSREM (top row), MR guided BSREM using the anatomical priors without motion correction (middle row) and MR guided BSREM with motion correction (bottom row). MR guided BSREM shows better SNR and higher image resolution compared to the BSREM method and as shown by red arrows (e.g., pituitary gland and right occipital & parietal cortex), incorporating motion correction into MR guided BSREM (bottom row), results in a sharper image; however, because of the lower counts on this exam, less improvement is observed.

Figure 4: The motion corrected T1 maps show an increase in image quality. Respiratory blurring especially at the dome of the liver and around blood vessels could be reduced yielding similar image quality than the breathholdscan. Although the motion model was built for the liver, also the visualization of the kidneys is improved. It should be noted, that the breathhold data cannot be directly compared to motion corrected images, because they were acquired in two separate scans.

Figure 1: Calibration: The M-sequence S_Tx gets transmitted by a sending antenna Tx and correlated with the received response S_Rx to create an impulse response R_xy. The principal components of R_xy get linearly fitted to the registered changes in a selected region of interest in the dynamic scan. Correction: Based on the motion model, radar signals obtained simultaneously to the T1 mapping sequence are used to predict respiratory motion shifts which can then be utilized during image reconstruction to correct for motion artefacts.

Figure 2. Experiment setup with the accelerometer placed on top of the abdomen.

Figure 5. 3D view of data binned by k-space center (left column), accelerometer (middle column) and Pilot-Tone RF transmitter (right column). Data binned by the three methods were found to be in good agreement. Additionally, data binned using the accelerometer signal showed fine tissue boundaries (green arrow) and data binned using Pilot-Tone showed finer liver anatomical details (yellow arrows).

Orbital navigator k-space trajectory. Left: Parametric plot of the trajectory shape. Right: Plots of the trajectory, gradients, and slew rate over time. The trajectory is made up of three orthogonal circles, with smooth transitions in between them. At a radius of 200 rad/m, the navigator gradients can be executed in about 1.65 milliseconds.

Sequence diagram of a 3D T2*-weighted FFE sequence with 3D orbital navigator gradients inserted after the excitation and before the phase-encoding gradients.

Figure 1: Examples of the simulated navigators. Water navigators left, fat navigators to the right. Fat navigators were not simulated for echo times of 10 ms and larger, because of the short T2* of fat.

Figure 5: The fit value (see figure 4) at 10 mm motion score of the golden standard, for all navigators simulated. For each resolution the readout durations (RO dur.) and echo times correspond to (from left to right): No EPI, EPI SENSE 5, EPI SENSE 3, EPI SENSE 1.