Fig. 1. A. Experimental set up. B. Acquisition protocol: dots show the sampling scheme within hardware (black), cardiac (red) or PNS (blue) limits for the Y-axis. C. Descriptive statistics of the measured relative distances of body landmarks. D. Contour plots of the absolute value of the magnetic field for a 300 mT/m gradient. The approximate positions of the eyes for prostate, heart or head landmarks are shown. Top row: |B| for a 300mT/m X-gradient (Y=0). Bottom row: |B| for a 300 mT/m Z-gradient (Y=0).

Fig. 3. 'Magnetophosphenes' occurrences for three imaging landmarks and three sampling axes. The dots are colour-coded according to the percentage of reported effects. The hardware limits are depicted as solid black curves, the cardiac stimulation limits as red dot-dashed lines, and the PNS limit as blue dashed lines, for X, Y, Z axes separately.

Figure 5: PNS hot-spots in the female model expected for different body positions in the coils B1 (top) and B2 (bottom). Colored spheres show hot-spots with largest PNS oracle (smallest PNS threshold). For abdominal imaging, we show both head-first and feet-first supine poses; all other scan positions use either head-first (head/cardiac imaging) or feet-first pose (pelvic/knee imaging). In all cases, meant to correspond to conventional clinical patient positions, the optimized coil retains some value in raising PNS thresholds, except for the head-first supine abdominal imaging.

Figure 1: A: Definition of the female model Huygens’ surface populated with 2497 magnetic basis functions. B: E-field induced by single basis near the heart. C: After performing E-field simulations and extracting PNS oracle values for all basis functions, we assemble the Huygens’ P-matrix linking all 2497 bases (columns) to PNS responses of all nerves (rows). D: PNS oracle for single 0.1 mm segment of the cauda equina mapped onto the Huygens’ surface (highlighted row in P-matrix). E: PNS oracle of all nerve segments for a single Huygens’ basis function (highlighted column in P-matrix).

Figure 3 – Color map of the normalized electric field produced by the z-directed magnetic field at 1000 Hz on a sagittal section of the leg. The color scale is arranged to fit the maximum value, in the presence of the implant, on this slice. The voxels that do not take part in the analysis are colored in black.

TABLE 1: Spatial peak of electric field, induced by a x-, y-, or z-directed magnetic flux density of 1 mT at 1000 Hz, in the absence/presence of the knee implant and the corresponding value of magnetic flux density which brings to the limit recommended by IEC 60601-2-33 for normal operating mode. The tissue where the spatial peak of electric field has been detected is indicated between brackets.

Figure 3: Nerve membrane responses in terms of transmembrane potential difference (2nd row) and sodium/potassium permeability (3rd row) for a symmetric and two asymmetric waveforms. All waveforms achieve the same k-space coverage. The symmetric waveform causes three APs. Using an asymmetric waveform (case A) with longer rise time t₁ (up-ramp) and shorter rise time t₂ (down-ramp) reduces the nerve’s depolarization and increases it’s hyperpolarization, thus suppressing AP formation. The asymmetric waveform case B has the opposite effect, inducing even more APs.

Figure 5: PNS thresholds given as gradient amplitude (left) and corresponding k-space coverage (right) for the parameter space spanned by the independent rise times t₁ (y-axis) and t₂ (x-axis). Each square corresponds to a T = 1 ms period waveform. Waveforms along the diagonal (white-dashed line) correspond to symmetric waveforms, while all other waveforms are asymmetric. The asymmetric waveform (A) achieves a 38% increase of PNS threshold in terms of k-space coverage, compared to the symmetric waveform S. The threshold of the asymmetric waveform B is 41% lower than that of waveform S.

Fig. 1. Schematic illustration of the segmentation pipeline to obtain a subject-specific body model for RF exposure analysis. The semi-automatic segmentation process involves many steps with elaborate user interaction, while the deep learning approach is able to generate an accurate body model directly from 7T T₁-weighted images.

Fig. 4. Comparison of simulated SAR_10g distributions in ground truth (top) and network-generated body models (middle), and corresponding difference maps (bottom). Figure headings denote peak SAR_10g values (top, middle) and relative difference of the peak SAR_10g (bottom).

Figure 4. Comparison of full pTx MPRAGE (inversion and excitation) with the commercial coil (left) and self-built coil (right) for different subjects at different scanners. Both images were acquired with FOCUS inversion pulses and UP excitations designed for their respective coils and are windowed the same. The table lists calculated, scanner-predicted, and scanner-measured local SAR in W/kg. The image quality is comparable, but the estimated SAR is lower with the self-built coil and EMF-based VOPs.

Figure 2. Local SAR estimates for 10,000 random unit-norm shim configurations as well as CP and each VOP model’s “worst case”. Top row) local SAR distributions for the commercial coil with constant safety factor VOPs (blue) and the self-built coil with EMF-based VOPs (purple); the distribution mean, standard deviation, and skewness are listed as well. Bottom row) zoomed in distribution of self-built coil local SAR.

Figure 4: Generalization Analysis: Ground-Truth local SAR distribution, predicted local SAR distributions ($$$\widehat{\mu}$$$), absolute error and estimated uncertainty ($$$\widehat{\sigma}$$$), for: A body array with 8 fractionated dipoles placed on the generic body model Duke¹⁰ (without body profile deformation) for prostate imaging (A) and liver imaging (B); Three head arrays, one with 8 fractionated dipoles (C), one with 8 oblique fractionated dipoles¹¹ (D) and one with 8 rectangular loops¹² (E).

Figure 1: Six example results of the 3-Fold Cross-Validation: Ground-Truth local SAR distribution (first column), predicted local SAR distributions ($$$\widehat{\mu}$$$, second column), absolute error (third column) and estimated uncertainty ($$$\widehat{\sigma}$$$, fourth column). On top are reported the peak local SAR (pSAR_10g), the root-mean-square error (RMSE) of the absolute error, and the root-mean-square (RMS) of the estimated uncertainty.

Figure 5. RF shimming inside the indicated ROI using the scanner’s shimming toolbox vs. the IF modes. Turbo spin-echo images were acquired, and the electrode’s tip temperatures were measured for both shimming scenarios.

Figure 1. Summary of the proposed method to obtain the implant-friendly modes of a TxArray coil.

Figure 1: Geometry of (A) 1.2 T high-pass radial planar birdcage coil and (B) 1.5 T high-pass birdcage coil. (C-D) B1⁺ field maps on the central coronal and axial planes passing through the human body model with no implants. The input power of the coils is adjusted to generate a mean B1⁺ = 4 μT on a circular axial plane passing through coil’s iso-center. (E) Example of mesh distribution in the lead, insulation and SAR box. The distribution of 0.1g-SAR in a plane passing through the lead tip has also been shown for a patient with bilateral leads.

Figure 3: I Picture showing anthropomorphic phantom in 1.2T vertical scanner (left) an in 1.5T horizontal scanner (right). II DBS lead attached to temperature probes implanted into the skull. III Schematic of phantom setup showing different trajectories (A, B, C and D) with different loop position. The red dot represents the 40cm mark measured from intracranial end, representing the end-point for 40 cm wire.

Figure 3: Sagittal views of (a) Pre-procedure 7.0-Tesla T2-SPACE imaging in the unpreserved human specimen shows high conspicuity in Meckel’s cave and the trigeminal ganglion. (b) The same location in the post-procedure imaging reveals widespread hypointensities.

Figure 2: Coronal views of (a) Encephalomalacia on T2-TSE imaging and (b) nerve root atrophy on CISS imaging seen in a patient with four prior rhizotomy procedures at 7.0-Tesla.

Figure 1: Magnetic stimulator. A) The pulse generator consists of a capacitor bank (C=110 µF total) that is charged with a high-voltage power supply. When the relay is closed, the capacitors are discharged into a coil, creating a damped sinusoidal coil current. The capacitor charge/discharge can be controlled remotely. B) RCL circuit diagram for the capacitors, relay and the coil. C) Flat pancake coil (D_out=40 cm, D_in=2.5 cm, 38 turns of 2.3-mm copper wire) and plexiglass table on which the pig is placed. The coil is attached below the table and centered below the porcine heart.

Figure 2: Simulation workflow. A) We acquired multi-bed Dixon images to generate porcine voxel models including 11 different tissue classes assigned with electrical conductivity values. B) We used CINE images (diastolic phase) to create 3D heart models to which we added cardiac Purkinje and ventricular muscle fibers networks. C) We simulated electric fields induced in the porcine models by a 40-cm pancake coil. D) We projected the electric fields onto the cardiac fibers and predict cardiac stimulation thresholds using electrophysiological models of the cell membrane ion dynamics.

Figure 3. The sites of stimulation of the human-body model for the “Aera” Gy gradient coil. Here, the head of the model was in the ROI.

Figure 4. Maximum intensity projection of electric field magnitude produced by “Aera” Gx (a), Gy (b) and Gz (c) gradient coil for the Yoon-Sun human-body model with a position of a radiologist. Here, the slew rates of the gradient coils are set to 100 T/m/s.

Figure 3: Measured and calculated PNS thresholds as well as hardware limits for seven gradient directions of H3, shoulder-coil contact body position. Individual PNS threshold measurements are shown using open markers. Logistic regression mean experimental PNS values are shown with solid (filled) markers and error bars. Linear fit to measured PNS thresholds are shown as solid lines. Hardware limits are shown as dashed lines. Calculated PNS thresholds shown as dotted lines. Chronaxie value of 600 µs was used for the head gradient calculations. Vertical axes represent peak to peak ΔG.

Figure 2: Calculated E-fields per unit slew rate on the surface of 50^th percentile male body model for (a) asymmetric GE head gradient, (b) symmetric H3 head gradient (in contact with shoulder) and 40 x 20 cm elliptical cylinder for (c) BRM body gradient coil. E-field directions and peak E-field value are shown. Only negative Z coordinates for BRM coil are shown due to symmetry. Peak E-fields occur at the face and top of the head for X and Y gradients of the asymmetric GE coil, R/L side of the neck and shoulder region for the Y gradient of H3 and R/L side of the torso for the BRM coil.

a) Photograph of PNS measurement setup. b) Coil design, consisting on two pairs of coils. We define the center of the coordinate system to be at the mid-point between the centers of the coil pairs (the z-axis is shown for further reference). c) Magnetic field strength distributions for homogeneous (left) and inhomogeneous (right) configurations, simulated in Comsol for a current of 1 A running through the coils, and plotted for the x = 0 plane, with 82 mm gap between plates. The structure along the contours is due to finite numeric precision in the simulations.

Experimental PNS threshold determination of sinusoidal (left), triangular (middle) and trapezoidal (right) trains of 1,000 pulses on 51 volunteers. Blue circles are semi-transparent and represent the threshold for every individual. Orange squares denote the mean for a given waveform and time parameters. The dashed orange lines are fits to standard models [6], with rheobase 30.3 ± 0.3 mT (30.4 ± 0.5 mT) and chronaxie 770 ± 110 us (519 ± 53 us) for the sinusoidal (triangular) scans. The red curves in the scans with linear magnetic ramps correspond to the IEC default thresholds.

Figure 1: 2D view of the sWBG (A) and aWBG (B) gradients showing the increase in diameter of the bore outside the field of view (FOV) for the aWBG gradient coil. The diameter of the FOV in the transverse and axial dimensions are depicted by the blue rectangle.

Figure 3: The primary (blue color) and shield (red color) coils of the whole-body sWBG (top row) and aWBG (bottom row) designs are shown in 3D for the x, y and z gradient coils. Notice the axial asymmetry of the aWBG transverse coils compared to the sWBG transverse coils. P indicates the patient end.

Comparison between simulated (left) and measured (middle) surface temperature after 6min permanent diffusion encoding with a b value of 101735 mm²/s. (I_max=100A,I_Eff≈50A, t_R=200ms). Image of the actual coil (right). Note that left side of actual coil is not included in simulation since it does not contain any thermal active components.

Example for time dependency of the current in the diffusion weighted local gradient coil. The red line indicates the corresponding effective current of the sequence which would emit the same amount of power within the coil. Sequence parameters are: t_R=2500ms, Slew-rate=2000T/s, Gradient-Efficiency=300mT/m/A, b-values ~ 3280s/mm², 14700s/mm², 74900s/mm²

Figure 1. The dreMR insert coil (white) placed within our 0.5T cryogen-free superconducting magnet, so that their isocenters coincide. A gradient coil (green) is present but was not used during this experiment.

Figure 2. (TOP) Temperature profiles for multiple coils within the superconducting magnet. (BOTTOM) Temperature profile for our shield thermocouple within the dreMR insert. Each pulse corresponds to a set of parameters in Table 1. Noise in superconducting temperatures increases during dreMR pulse sequences. A black vertical line denotes where a prolonged break was taken before returning to the experiment. No pulses were applied during the time that was removed.

Figure 1. An acoustic shield, made of ¼ inch thick clear acrylic, is placed over a GE Head Neck Unit on a PET MR table.

Figure 3. Sound pressure level of MRI acoustic noise measured when scanning volunteers in a GE 8ch high resolution brain coil with and without acoustic shield

Figure 1. (Left) Elevation view of the head-only RF patient-bore structure. For the structure made from FRP, a stainless-steel mesh RF shield was mounted on the outer-diameter of the structure. In the case of CFRP, no stainless-steel mesh was needed as the CRFP material acts as an RF shield at 128 MHz. (Right) Experimental setup for measuring sound pressure levels. Here the patient-bore structure has been inserted into the gradient, and acoustic measurements are made at imaging isocenter, i.e., the narrow portion of the shield inner bore.

Figure 2. Results of the A-weighted, equivalent continuous sound level (LAeq, dBA) tested for 11 clinical sequences or sequence variations using the Fiberglass RF shield structure (black bars) and Carbon Fiber RF shield structure (gray bars). Shown with the limits assuming hearing protection with noise reduction rating (NRR) = 26 and 33 dB to reach the U.S. Food and Drug Administration (FDA) and IEC 60601-2-33 (3rd Edition) guidelines for nonsignificant risk limit of 99 dBA.

Table 1: SAR prediction results of 3T and 7T images

Table 2: Training performance

Motion study B1 and SAR maps for a randomly selected position.

Pose study across all perturbation states for the estimated values (blue) and the true values (red). Top left: head averaged SAR, top right: peak local SAR, bottom left: mean B1+, bottom right: standard deviation of B1+.

The blue points indicate the SAR that would be estimated from an unperturbed model while the red points show the ‘true’ SAR from the true model.

Fig. 2. Comparison of multiple excitation scenarios. Patient centric coordinate system with origin at heart. Simulations of Ella are only available in the light gray area. A) Amplitude limit of all simulations. B) Maximum normalized SAR in each group for cases of Fig. 1B. C) Overlay of the voxel models. D) Ratio between optimized mean(B₁⁺) of image slice (i.e. excitations of same or better homogeneity than CP mode) and mean(B₁⁺) of CP reference mode. E) same as D) but multiplied with safety factor obtained from the respective (i->j) maximum normalized SAR (CP reference mode unchanged).

Fig. 4. A) Sweep analysis of SAR-limit (SL, black) and amplitude limit (AL, blue) of human voxel model Duke in center position using interpolation from two anchor simulations shifted by ∆z=±50 mm (cf. Fig. 3). 10000 random excitation vectors are scaled to satisfy the anchor limits. A maximum normalized SAR > 1 (red) is unsafe, a normalized SAR < 1 (gray) wastes B₁⁺ performance. AL is more conservative than SL. B) Corresponding curves if extrapolation from only one anchor simulation at ∆z=+50 mm is used.

Figure 2 (a) SAR_10g distribution when d=15mm in standing position. (b) SAR10g distribution in supine position. (c) SAR_10g distribution when d = 225 mm in standing position. (d) SAR_10g distribution (exclude hands) when d = 225 mm in standing position. (e) SAR10g distribution with a conventional whole-body coil in supine position. (f) SAR_10g distribution (exclude hands) with a conventional whole-body coil in supine position. Legend is individually adjusted for best visualisation effect, the explicit value can be found in Table I and II.

Figure 1. (a) Duke model in the MRI-Linac coil in the standing position. The distance from the top rung to the ground is 187cm. The variable d denotes the distance from the bottom rung to the genital area. (b) Duke model in the MRI-Linac coil in supine position. (c) Duke model in a standard 32-rung birdcage coil.

Fig. 1 (a) Simulated model (at 320-mm shifted distance) on the electromagnetic analysis software. The blue dashed line represents the center of the MRI scanner and the RF transmission coil. The blue arrows show the direction of human phantom shift. A white arrow head shows the contact point of the right elbow on the bore wall. (b) The SAR_10g at the contact point versus the shift distance.

Fig. 2 The electric field distributions of (a) an empty scanner, (b) the position of contact point with 320-mm shifted distance, (c) the position of contact point at the center (0 mm). The white arrows show the contact points. The red dashed lines represent the projection of the birdcage transmission coil elements onto the plane of each map. To improve the visualization of the electric field strength near the human model, the color index was maximally scaled at 2000 V/m.

Fig. 1. The proposed multi-task learning model. The generator loss newly incorporates perceptual loss, local SAR min/max loss, and spatial overlap loss using combined Tversky and focal loss function¹⁰ to penalize overlap errors in the higher SAR. The discriminator loss includes consistency loss between outputs from the real SAR and its AugMix¹¹ version to improve the robustness to data shift. The model parameters are initialized with He normal¹² and optimized with ADAM.¹³ Learning rate is 0.0001 and reduces after 40 epochs. The size of mini-batches is 8, and the number of epochs is 100.

Fig. 5. Visualization examples of local SAR prediction on a specific case of (a) Duke, (b) Ella, and (c) Louis head models. For simplicity, the center slices in each axis are visualized. Overall, local SAR distribution is comparable to ground truth SAR. We observed discontinuity between slices in the SAR prediction with peak values estimated slice by slice (blue). Bone and air information was predicted due to tissue properties data.

Fig. 2. (a) The coil loaded by two phantoms, i.e., PVP-based with electrical properties close to human brain and oil-based phantom with small losses, as well as by human model. (b) Birdcage coil overview. The scanner inner bore diameter was 560 mm; total coil length was 450 mm. The coil was shielded by a metal enclosure with inner diameter of 610 mm and length of 1590 mm. (c) Close-up cross-sectional view of the high resolution human head model.

Fig. 1. Reverse engineering workflow.

The top row shows the dielectric of Duke's brain and the implant that is added. The row below shows the electric field components as computed by FDTD. Using the current density profile that results from the inverse calculate from Equation 1 as input for the function $$$f(\cdot)$$$ we obtain the electric field as shown in the third row. Finally, the bottom row shows the difference between the field components. All the fields are shown using the same arbitrary unit scale.

Equation 1 is solved for a small copper implant using the proposed method. The resulting current density profile is compared between the offline (using $$$Z$$$) and online (using $$$f(\cdot)$$$) versions of this method. The background conductivity and the conductivity of the copper cube are shown in the bottom row.

Fig. 2. Phantom measurements without and with the high permittivity pad in place, with and without insulation. Shown are transverse cross-sections of B₁⁺ efficiency (top) and temperature increases (bottom). The position of the high permittivity pad is illustrated in white, and the numbers denote the maximum temperature increase.

Fig. 1. Simulated effects of high permittivity material positioned on the anterior side of the male body model, with and without insulation. Shown are transverse cross-sections of the B₁⁺ efficiency (top) and coronal maximum intensity projections of SAR_10g (bottom). SAR data were normalized to an average B₁⁺ magnitude of 1 uT within the transverse cross-section of the body model. The position of the high permittivity material is illustrated in white.

A few large floating cable traps placed on a cable connecting an RF coil to the main system. A portion of the cable is selected, comparing the standard floating trap design to caterpillar traps. Standard floating traps inductively couple to the shield current, $$$I$$$, and impose a large impedance. In contrast, our distributed design of caterpillar traps consists of smaller toroids each inductively coupling to the shield current. Although each toroid imposes a smaller impedance, by covering the full length of the cable with these traps, shield currents can be sufficiently attenuated.

B1 mapping experiments. Initially, a B1 map with 30 degree target flip angle of a slice 1cm below the top of phantoms is acquired with three configurations and displayed in 0-65 degree range: a) phantom only, conductor equipped with b) spacers only and c) caterpillar traps. d-e) Results from a) and c), respectively, displayed in 22-37 degree range. To test the performance under stress: f) a smaller non-loading phantom, g) a larger loading phantom are placed as weight on the caterpillar traps and the maps are shown in 22-37 degree range. Putting weight on the traps has minimal effect.

Fig. 1 Diagram of a first-order Lattice balun (left) and the novel LCCC balun (right).

Fig. 5 Bench test measurements of the common-mode rejection ratio (top) and insertion loss (bottom). When measuring the common-mode rejection ratio, the current probes’ free space the response was first calibrated to 0 dB as a baseline.

Figure 1: CSI images of the thermal ¹³C Urea phantom (a, cryocoil), (b, surface coil) and (c, volume coil). It can be seen that both surface coils suffer from an inhomogeneous transmit and receive profile. (d), (e) and (f) show spectra from the voxel indicated by the black box in the above images. The spectral range indicated by the arrows was used for the calculation of noise distribution. (g), (h) and (i) show the the Rician noise distribution from the range indicated in the above spectra.

Figure 2: The peak intensity maps for the cryocoil (a) and surface coil (b) at RF powers that yield similar excitation profiles for both coils(c). (d) and (e) show an example of the noise maps, which were determined by fitting a Rician distribution to the noise region shown in Fig.1. The SNR ratio along the lines in (a) and (b) is shown in (f). (g) and (h) show the calculated SNR maps of the cryocoil and the surface coil. In (i), the ratio of the two SNR maps is shown.

Fig. 4. (a) Photograph of the array with cover removed; (b) measured SNR of the element as a function of edge-to-edge distance to the neighbor element; (c) photograph of the imaging setup.

Fig. 3. Impedances at the input terminals of the matching and decoupling circuit (normalized to impedance of the coil Z_c).

Figure 3: (Left) Simulation model of the birdcage coil (with tuning capacitors labelled) used to compare performance of the (Middle) simulation model of the metamaterial resonator with human body model. (Right) Photograph of the constructed metamaterial liner resonator. Small wires (0.6 mm diameter) join the C^r capacitor pads to the aluminum mesh through contact between conductive rails that press against aluminum tape on the inner side of the polycarbonate shell.

Figure 4: (Left) Simulated transmit performance of the metamaterial liner and single tuned birdcage coil for ¹H – 200 MHz and ²³Na – 52.9 MHz imaging. (right) Maximum intensity projections of the local 10g average SAR normalized to a mean 1μT transmit field in the region outlined in the transmit efficiency maps for ¹H–(200 MHz) and ²³Na–(52.9 MHz).

A – ¹H image of a 3-section phantom filled uniformly with aqueous NaH₂PO₄ solution, B – ²³Na image of a 3-section phantom. Section 1, “M” contains aqueous NaH₂PO₄. Section 2, “R”, contains a solution of NaCl and MnCl₂. Section 3, “I”, is filled with the MnCl₂ solution.

A – ¹H butterfly-type coil model with tuning/matching capacitors. B – X-nuclei tuneable coil model. Feeding of the X-nuclei coil is provided via inductive coupling to a small loop coil.

Figure 1: Surface coil geometry.

The geometry of the (A) circular, and (B) fractal coil. Each were copper etched on a 100x100mm FR4 substrate with breaks in loop for the tuning capacitors (C_s1-4) and the matching capacitor (C_p). One end of the loop for both coils was connected to the T/R switch via a coaxial cable while the other end of the loop was grounded using the same coaxial cable to form a one port network.

Figure 3: B₁⁺ Field Maps.

The B₁⁺ field maps for the (A,C,E,G) circular coil, and (B,D,F,H) fractal coil. A and B show the sagittal field map and C-H show the three coronal maps which are slice centered at depths of (C,D) 15mm, (E,F) 25mm, and (G,H) 35mm.

Fig. 3. SNR maps of 1-, 4-, and 28-channel coils at cryogenic and room temperatures. The SNR has an arbitrary unit. The peak SNRs are 247 for 1-channel cryogenic coil, 130 and 296 for room-temperature coils separated from the phantom by 22 mm and 2 mm; 417 and 270 for 4-channel cryogenic and room-temperature arrays separated from the phantom by 13 mm; 265 for 28-channel array. The ratio of cryogenic SNR to room-temperature are 1.90 for the 1-channel coil and 1.54 for the 4-channel array.

Fig. 4. Cross-section SNR profiles of imaging experiments. The SNR has an arbitrary unit. The SNR values are averaged between two nearest pixels. The inset shows the positions of cross-sections.

Figure 4. Three representative transversal slices of ³¹P SNR maps of Pi acquired at (A) 7T and (B) 10.5T, overlaid with 2D CSI slice (extracted from 3D data) acquired at the 90^o nominal flip angle of the global Pi signal. The noise level was zoomed in ×1000 times along the vertical scale. This figure clearly shows significant improvement in spectral quality and SNR at 10.5T.

Table1. The parameters and their ratios measured at 7T and 10.5T and used for quantification.

Figure 1 Photography of the first 9.4T whole-body (830 cm diameter) ultra-high field MRI in China (a), the nested dual-tuned proton-sodium multi-channel loop-array transceiver coil (b) and the RF interface (c).

Figure 4 Proton（left）and sodium（right）images obtained by the dual-tuned coil array at 9.4 T.

Figure 2. Fabricated flexible 32-channel ¹³C receive-only array (f = 32.1 MHz), attached to the head phantom used for MR characterization. The projection of the 7 slices used for the volumetric characterization is also noted.

Figure 3. SNR maps of the seven slices covering the whole brain area of the head phantom, and comparison to the results measured with a reference birdcage coil.

Figure 1. A) EM simulation model of the double-tuned ¹³C/ ¹H array coil loaded by the HS phantom. B) Photo of the ¹³C/ ¹H array with the RF shield removed for better visualization. C) EM simulation model of the ¹³C/ ¹H array loaded by the Duke voxel model. In Figure 1B, dipoles and loops are marked by red and yellow dashed lines, respectively.

Figure 5. Central sagittal in-vivo human brain GRE images (A, B, C) and corresponding B₁⁺ maps (D, E, F) obtained using the presented double-tuned ¹³C/ ¹H array (A,D), double-tuned 20-loop ³¹P/ ¹H array (B, E), and single-tuned 16-loop double-row ¹H array (C, F).

Figure 1. Diagram of the dual tuned balun and approximations for different nuclei. The dual-band balun approximates two different single-band baluns at different frequencies, making it suitable for dual-band detection systems.

Figure 5. Common mode rejection ratio test with current probes. The CMRR shows a good rejection ratio at sodium and proton frequencies. The two current probes are connected to the VNA while the ground of the two cables from the balun circuit connects to a large resistor.

Figure 1: Photographs of the coil (A) Coil inside of shield (B) Coil showing alternating rungs and match/tune configuration for the hydrogen port. (C) Close up view of the jumpers. The hydrogen coil can be identified by the greater number of breakup capacitors on the legs, the capacitors on the end rings bridging the sodium legs, and the shorter length compared to the sodium coil. (D) Cross-sectional block diagram showing field orientations with respect to the legs and the respective feed ports.

Figure 3: Image data. (A) Birdcage Transmit/Receive with loop terminated in 50Ω load. (B) Birdcage transmit, loop receive for hydrogen. (C) Birdcage transmit,loop receive for sodium

Figure 4. (a) Measured sensitivity profiles of the individual array channels for ¹³C (top) and ²³Na (bottom). (b) ¹³C Noise correlation matrix. (c) ²³Na Noise correlation matrix.

Figure 1. Fabricated flexible 8-channel ¹³C receive array (f = 32.1 MHz): a) with the 8-elements visible sewn into a flexible cloth, b) covered with a flame-retardant protective cloth and, c) wrapped over the head-size phantom used for imaging. All the electronic components (matching network, active decoupling, LNA) are integrated into one PCB, and enclosed into an ABS box (60x35x15 [mm]).

Figure 3: B1+ field simulation results of the dipole coil over a homogeneous phantom and a heart numerical model.

Figure 5: Left: cardiac mid-short axis localiser and CSI grid. Right: mid-septal 31P cardiac spectrum (apodised by 30 Hz) from the voxel highlighted in the CSI grid.

Figure 2. Comparison of single-loop-coil positioning with rat between the 2D- and 3D-shaped. The 3D-shaped coil base was 3D printed out that fits better to brain geometry than 2D flat coil. The gap between coil and skull will be muscle and skin that are not shown in CAD.

Figure 3. Comparison of image qualities of rat brain between 2D- and 3D-shaped loop coils. Both have good whole brain coverage but relatively the 3D-shaped loop that fitted better to brain geometry exhibits higher SNR and better field homogeneity in T2-weighted and EPI images and Field maps. The 3D-shaped coil may have more efficient coverage on most brain areas with nearly uniform field signals based on field biases from T2 weighted images (good homogeneity highlighted in red frames, prefrontal difference pointed out in arrows on EPI).

Figure 1. Designs of the coils (left: 26 mm, right: 64 mm diameters; D26 and D64 coils, respectively).

Figure 3. Brain specimen images of a Alzheimer’s disease patient using the D26 coil in isotropic 50 μm resolution (left: axial, right: coronal). Numerous tiny dots were considered to be iron-loaded amyloid plaques.

The design of small surface coil (a), saddle coil (b) and surface coil combined with the saddle coil (c and d). The diameter of all copper wires is 1 mm.

(a) The position of 12 selected ROI overlaid on T2WIs using small surface coil only (left) and proposed coil (right) respectively. One group of ROIs (1 to 4) is located at regions close surface coil, another group (5 to 8) is located at regions with middle distance from surface coil and the other group of ROIs (9 to 12) is at regions with farthest distance from surface coil. (b) The SNR results from 12 ROIs using small surface coil only and proposed coil respectively.

Fig. 5 MR microimaging results of the human embryo acquired with the 6-turn and saddle coils. (a-d) Sagittal views. (e-h) Axial views. In Figs. (e)-(h), the magnified images near the trigeminal fibers (red ovals) and the border of the medulla oblongata (blue ovals) were shown on the right side of the original images. NEX: number of excitations.

Fig. 3 Simulated SNR maps. (a,b) Intrinsic SNR maps for (a) 6-turn DHD and (b) saddle coils. (c) Relative SNR of the 6-turn DHD coil with respect to the saddle coil.

Figure 1: Block diagram of the standalone wireless automatic tuning and matching system

Figure 3: Return Loss data a) Tx coil b) Rx coil with measured S-parameters c) Rx detuning during transmit

Figure 1: The fit of the bespoke coil and the animal chair (A), a macaque in the coil (B), and a schematic showing the receiver loops (C)

Figure 3: Transverse, coronal and sagittal views of the SNR (first three rows) and tSNR (last three rows) maps of the head for in-vivo scans acquired using the 4-channel flexible coil (first column) and the 15-channel coil (second column). The improvements are most pronounced in the frontal regions, however, a meaningful increase in imaging performance can be observed through the brain. Note that stock reconstruction was used, and the images were not brain extracted.