Fig. 5 Heavily T2-weighted fluid sensitive imaging of the brain shows good suppression of white and gray matter, with bright CSF in the subarachnoid space. Shown is A) the phase difference image, B) the magnitude of the RF phase modulated GRE image, and C) the cross product image that appears as a heavily T2-weighted image.

Fig. 2 The signal for two data sets $$$(-\Delta\phi$$$ and$$$+\Delta\phi)$$$ is shown in the complex plane for Bile ( in green) and Liver ( in red) tissue. $$$\theta_{BG}$$$ is the background phase. The cross product of the two vectors that have a smaller angle ( $$$S_{Liver}$$$ and $$$S_{-Liver}$$$) will be smaller than the bile $$$S_{Bile}$$$ and $$$S_{-Bile}$$$.

Figure 5: Comparing 32x32 ME-bSSFP and CSI data arising from a large pancreatic tumor in a mouse at ~2 hrs after injection of deuterated glucose. The regional distribution of the metabolites is similar (color, superimposed on 1H anatomical images in grayscale), with glucose located in the bladder, lactate in part of the tumor, and water diffusely distributed. (B) CSI data for all pixels, with lactate visible on the spectrum. C) SNR comparisons for matched ROIs. Glucose SNR was tripled with bSSFP vs. CSI (57 ±30 vs. 19±11, p<0.001, N=10). Water SNR was doubled (13±5 vs. 7 ±3, p=0.005, N=8).

Figure 4: Phantom study with 3 tubes containing deuterated water, glucose and lactate. (A) ME-bSSFP isolation of the different metabolites, with some cross-talk. (B) ²H NMR spectra summarizing the CSI data for all spatial elements, showing the different metabolites. (C) Summary of the SNR measurements showing significant improvements for lactate and glucose. The “avg bSSFP SNR” was estimated as the SNR obtained by simple signal averaging over the metabolite’s region (i.e., with no IDEAL processing).

Figure 1. Principles of rNOE saturation transfer during electrostatic binding of a small molecule to an immobile medium with charged groups (here -SO₃^- ). The aliphatic protons of free arginine are efficiently labelled by a radio-frequency (RF) pulse, and there is no rNOE based transfer within the molecule, which is in fast-tumbling mode. Bound arginine molecules are in the slow-tumbling mode, where there is efficient intramolecular rNOE transfer from aliphatic protons to exchangeable protons, and finally to water via chemical exchange processes.

Figure 2. CEST MRI of arginine solutions mixed with ion-exchange medium (with functional group -SO₃^-). (A) The chemical structure of arginine; (B) 1D ¹H NMR of arginine (5% H2O/95% D2O, pD 7.2), peak assignments are based on previous studies.^6,7 (C-D) Z-spectra of ion-exchange resin medium alone (C), medium with arginine (100 mM, pH of 7.2) (D).

Figure 3. Receiving operating characteristic (ROC) analysis showed the performance of ITSS combined with R2* in evaluating HCC histological grading, with an AUC of 0.856, sensitivity of 88.89%, specificity of 69.44% .

Figure 2. A 62 year-old male with poorly-differentiated HCC in the right lobe of the liver. (a) T2WI image; (b) phase map; (c) tumor was delineated around the edge of the tumor; (d) AS software recognized quantitatively and automatically ITSS ratio by reading phase maps. ITSS recognized were covered in green.

Figure 2. Comparison of T₂* maps before and after the B₀ field inhomogeneity correction using the proposed method. Signal loss from dephasing in the frontal region was successfully recovered.

Figure 4. Multi-modal images including MPRAGE+C, FLAIR, T₂', choline to creatine ratio and NAA to creatine ratio maps of a glioma patient. Concurrent increase of T₂' value and Cho/Cr and decrease of NAA/Cr in tumor area could be observed.

Figure 2. Pre-ferumoxytol 3D STIR-FSE multiplanar reformatted (MPR) maximal intensity projection (MIP) (a) demonstrates venous contamination obscuring the suprascapular (green arrows) and axillary (red arrow) nerves. On the post-ferumoxytol MPR MIP (b) the suprascapular (green arrows) and axillary (red arrow) nerves are clearly delineated.

Figure 1. Simulations of different relative doses of contrast agents (CA), for (a) Gadolinium+blood to muscle contrast, and (b) Ferumoxytol+blood to muscle contrast, at two different blood velocities, assuming TE=80 ms and TI=255 ms.

Figure 4: Resulting temperature maps for a cervical cancer patient during mild RF-HT of the tumor. Particularly for tumors close to the intestines, gas motion causes severe artifacts. The black arrows point at residual phase errors after background field correction, that appears to be less in the TFI. Furthermore, the LBV method resulted in the loss of valuable pixels. The comparison of the corrected temperature with a sensor illustrates how severe the susceptibility artifacts were in the uncorrected DEGRE.

Figure 3: Bowel motion-induced susceptibility artefacts in volunteers at constant temperature and their correction with LBV, PDF and TFI. The temperature error maps before and after correction are displayed. In the cumulative error plots, the ratio is calculated between the numbers of voxels occupying the given value and less over all voxel counts. In contrast to the simulation results, as seen in Fig.1, the cumulative error plots (last row) indicate that TFI results in the least residual phase errors.

Figure 1. An overview of the proposed method, using the brain as an example. (a) is the sampling patterns of the PEPTIDE sequence in k-space. Each blade here is sampled within one TR. (b) shows the reconstructed magnitude maps of each blade. (c) is the temperature maps derived from phase maps of different echo times, using the PRFS method.

Figure 4. Temperature curves of two typical voxels in the regions of interest for four temporal cases. For most of the situations, the reconstructed temperature curves match with the reference with some mismatch in the TRs at the margin.

Figure 3. Single- and multi-baseline PRFS temperature maps at two positions of the respiratory cycle and plots of temperature change within an ROI (white circle) in simulation, phantom experiment with respiratory motion simulator, and in free-breathing acquisition in a healthy volunteer without heating. Temperature difference maps between two positions of the cycle (max-min) show that the multi-baseline reconstruction results in less motion artifacts compared to the single baseline reconstruction throughout the breathing cycle.

Figure 1. (a) schematic diagram of spiral-based MR GRE sequence and (b) real-time reconstruction pipeline. RF: RF pulse excitation, SS: Slice-Selection gradient, PE: Phase Encoding gradient, RO: Read-Out gradient. FFT: Fast Fourier Transform.

Figure 3. Representative results in a healthy volunteer. (a) Proposed calibrationless (calculated) B1 and measured B1 maps. (b) MPF maps estimated with the maps in (a). (c) Whole brain MPF histograms for three scenarios of availability of B1 information (measured, calculated, no B1 maps). Note high consistency of the results obtained with calibrationless and measured B1 maps.

Figure 4. Results of application of the proposed method to data acquired on a scanner from a different vendor. Note high consistency of the proposed and measured (AFI) B1 maps, as well as effects of MPF correction (as compared to no B1 MPF results). Application of the proposed method did not require recalibration of the relaxometric constants used in the technique.

Figure 3. Axial, sagittal, and coronal 3D k_PLBA maps as a function of changes in pyruvate and lactate excitation angle. (a) The 3D k_PLBA map for the case when θ_P = 10°, θ_L = 10°. (b) The 3D k_PLBA map for the case when θ_P = 20°, θ_L = 20°. (c) The 3D k_PLBA map for the case when θ_P = 30°, θ_L = 30°. (d) The 3D k_PLBA map for the case when θ_P = 60°, θ_L = 60°. (e) The 3D k_PLBA map for the case when θ_P = 10°, θ_L = 30°. (f) The 3D k_PLBA map for the case when θ_P = 20°, θ_L = 30°.

Figure 1. Axial, sagittal, coronal, and multi-plane 3D view of hand measured B₁⁺ field maps at ¹³C clamshell coil isocenter. (a) The axial view of the hand measured B₁⁺ field at coil isocenter. (b) The sagittal view of the hand measured B₁⁺ field at coil isocenter. (c) The coronal view of the hand measured B₁⁺ field at coil isocenter. (d) An axial, sagittal, and coronal multi-plane 3D quadrant view of the hand measured B₁⁺ field at coil isocenter. The red contour lines in all subfigures highlight the boundaries of the volume that produces deviations in the B₁⁺ field within ±10%.

Figure 4: In vivo relaxation quantification results. One exemplary slice for T₂ and T₂* maps from GESE-EPIK and the reference methods is shown in the top left corner. T₂* and T₂ histograms from all acquired slices compare each parameter distribution from GESE-EPIK with its reference method. Mean values of each distribution with its standard deviation are given.

Figure 2: Reconstructed images from 10-echo GESE-EPIK for four exemplary slices of an in vivo data set. The TE for each echo is given. The signal strength of later echoes is modulated for visualisation purposes.

Figure2: The APTw images of brain and phantom acquired by Full sampled 13 shots, CS 3 shots, CS 2 shots and CS 1 shots. CS 3 shots and CS 2 shots have similar APTw images as the full sampled one, but for CS 1 shot, some sharp boundaries were smeared or distorted.

Figure1: The k-space sampling pattern for 3, 2, and 1 shot compressed sensing acquisition.

Figure 2: Average $$$\delta\Omega^{2}$$$ values for four deep gray nuclei in 26 healthy adults plotted against estimated non-heme iron concentrations taken from Ref. 7. A highly significant linear correlation supports the hypothesis that $$$\delta\Omega^{2}$$$ is influenced by the presence of cellular-level iron. Error bars denote standard deviations.

Figure 1: Signals on a semilog scale from four ROIs - globus pallidum (GP), putamen (Pt), caudate (Cd) and thalamus (Th).

Figure 1: 3D Silent Parameter Mapping starts with a low FA~1° PD measurement (left) followed by a magnetization prepared segmented ZTE acquisition (FA=3°) to encode T1 and T2 contrast weighted information (right).

Figure 3: 3D Silent Parameter Mapping (FOV=19.2cm, res=1.2mm, BW=±31.25kHz, 1.5 averages, 6min05sec) at 1.5T showing PD (left), T1 (0…2s, middle) and T2 (0…1.6s, right) parameter maps.

Multi-slice layout of the T1 contrast images obtained from the T1/T2 prequence acquisition.

Multi-slice layout of the T2 contrast images obtained from the T1/T2 prequence acquisition.

Representative simulated field maps using both 8 and 6 phase-cycles (SNR=73). The PLANET field maps required phase-unwrapping while the proposed methods did not. The proposed methods in each case outperform PLANET field maps. The PLANET residual image contains artifacts due to the streak-banding caused by the linear off-resonance which can be seen in the estimate. PLANET with 6 phase-cycles performs poorly even in good SNR conditions using single or multiple TR-ellipses. Proposed methods result in good field maps, improving by an order of magnitude in almost every case.

Field map estimates from a cylindrical phantom. A modified TrueFISP pulse sequence with 200 dummy pulses ensured steady state. 8 linearly spaced phase-cycles between 0 and 2$$$\pi$$$ at TR=6ms and FA=70° were acquired for the single ellipse PLANET (left) while 4 linearly spaces phase-cycles were acquired for each of two TR-ellipses (TR1=6ms, TR2=12ms) (right). The PLANET field map estimate (left) suffers from phase wrapping and artifacts due to the high FA while the proposed multi-TR estimate only suffers from discontinuities due to the large difference between TR1 and TR2.

Figure 5: Dynamic MRSI of HP-[1-¹³C]Pyr and HP-[¹³C]-Bicarbonate in human heart from infusion rates of 2.0 and 5.0 cm³/sec. Image acquisition times following the start of infusion are shown below each image. The slower infusion allows HP-[1-¹³C]Pyr visualization in the heart right ventricle (RV) at the earliest feasible imaging time point, but for the faster infusion HP-[1-¹³C]Pyr has mostly transited to the heart left ventricle by this time.

Figure 2: Dynamic simulations of the HP-[1-¹³C]Pyr MR signal intensity (= concentration x polarization) in the heart left ventricle based on the model shown in Figure 1 using infusion rates ranging from 1.0 to 5.0 cm³/sec.

Figure 5 - In-vivo results acquired at 3T showing the a) preprocessed input image, used as a starting point for both b) MEDI and the c) Recurrent Inference Machine (RIM). The MEDI results shows high contrast in the deep nuclei regions associated with higher iron content. However, the image gets smoothed by the spherical filter. The RIM result has some element from the MEDI image such as dark contrast in the myelin regions, yet remains similar to the input image to a large extent. Possibly, the network did not complete training, or the training data was unsuited to this problem.

Figure 2 - The RIM inference process for an axial example drawn from the testing data, along with each estimate from the iterative MEDI procedure. The label (ground truth) and input signal are shown in the left-most column. In both techniques the shapes become clearly defined with increasing iterations, as expected. The initial inference from RIM resolve many shapes, but with an incorrect susceptibility range (e.g. sphere, middle-left) which does not get corrected over inferences. MEDI on the other hand converges closer to the label data.

Figure 2: Representative Sagittal/ axial slices of susceptibility maps (QSM) and B₀ maps are shown for each volunteer with scanner only shim and with local shimming. Arrows indicate inhomogeneities and artifacts. For better visualization, in each case a zoomed-in window is also shown (dotted green rectangle) with higher contrast

Figure 1: The sagittal slices of the first echo magnitude image of the field mapping sequence for each of the three measured subjects are shown on top. In a-c histograms of the B₀ maps for each volunteer measured using the scanner-only shim (red) as well as the local shimming method (blue) are shown.

Figure 1: Microscopic field inhomogeneity was simulated using FEM with a simple model consists of water and LDs. A rectangular LDs were randomly placed in the water region with (2.5 mm)². The susceptibility of triglycerides of 0.61 ppm was used. Different sizes ((200 nm)²-(600 nm)²) of LD were assumed for the modeling. After meshing the model, the field map was calculated using FEM.

Figure 3: R2* values were plotted fat fraction with LD size of (a) (200 nm)², (b) (400 nm)², and (c) (600 nm)², respectively. The slopes of the linear regression for the plots with LD size of (200 nm)², (400 nm)², (600 nm)² were 1.20 (95% CI = 1.11-1.29), 0.931 (95% CI = 0.841-1.01), and 0.761 (95% CI = 0.67-0.85) while the intercepts for them were 18.2 (95% CI = 16.3-20.1), 10.1 (95% CI = 8.20-11.9), and 5.81 (95% CI = 3.60-8.02), respectively. The simulation results fairly agreed with (d) in vivo measurement.

Figure 3: Box-and-whisker plot of the T1ρ variability (standard deviation, log scale) in the cartilage derived from images without motion correction (none) and images corrected using three motion correction methods (rigid, affine, and elastic).

Figure 2: T1ρ (left, color bar: ms) and R1ρ (right, color bar: s^-1) maps of the cartilage derived after elastic registration.

Fig. 2: Results from patient #2, using product sequences (a-d) and the present MPME sequence (e-h). MPME images are shown for all sampled pathways and a single echo time (out of three). These MPME images are not meant to be read directly, but contrast translation into traditional formats as shown in (a-d) will become possible only when the number of scanned patients is large enough for a neural network to be trained. An evolving infarct (green arrow) and a benign hemangioma (yellow arrow) are seen with varying degrees of conspicuity in the various contrasts and slices shown here.

Fig. 1: Examples for four whole-head 3D contrasts generated through the neural translation of a single few-min MPME acquisition, from a prior healthy volunteer study: quantitative T₁ (a) and T₂ (b) maps in ms, and qualitative grayscale T₂-FLAIR (c) and T₁-weighted (d) contrasts. In addition to these four contrasts, MPME also readily generates susceptibility-weighted, proton-density weighted, T₂-weighted and MPRAGE contrasts, along with quantitative flip angle maps.

Figure 1: In the first row the susceptibility map obtained by the proposed algorithm is shown.The second row shows the ground truth and in the third row a susceptibility map calculated by STAR-QSM is shown.

Table 1: The values for the image quality measures achieved by the susceptibility maps obtained by the proposed algorithm and by STAR-QSM are shown in the first and second column. The third column shows the achieved ranking compared to all algorithms testes in the 2016 QSM challenge. RMSE is the root mean squared error. HFEN is the high-frequeny error norm. SSIM is the structural similarity index. The ROI error is the absolute value of the mean error in selected anatomical structures as in the 2016 QSM challenge.

Figure 1: Smoothness-optimization algorithm for estimation of field inhomogeneity map. Blue panel: Preparation entails setting the initial estimate of the field inhomogeneity map, smoothing the estimate, generating a residual phase map, and calculating the smoothness parameter λ. Red panel: The background field inhomogeneity map is iteratively updated by addition of incremental phase and λ recalculated, until it is minimized. Green panel: The corrected field map is generated by subtraction of the final estimated inhomogeneity map, with final residual correction applied.

Figure 2: Field maps for five subjects. Data were acquired with the following parameters: FOV = 20 x 20 cm²; slice thickness = 5 mm; flip angle =12º; TR/ΔTE = 25/ 6.12 ms; bandwidth = 62.5 kHz; 200 phase-encodings; 200 readout points; reconstructed matrix = 400 x 400; voxel size = 1 x 1 x 5 mm³. Reconstruction performed according to Lee et al. (4). Column 1: magnitude image; column 2: raw field map; column 3: field-map corrected using weighted least-squares method; column 4: field-map corrected using smoothness optimization method.

Figure 1. The architecture of the 3D U-net used for Background field removal (BGFR). The convolutional neural network (CNN) displays the output local field maps (B_int) the same size (192x192x96) as the input total field maps (B_tot) . The number of feature maps is shown on top of each of the network’s layers.

Figure 2. A set of training field maps inside the brain simulated in a head and neck phantom with random deformations. A sagittal and a coronal slice of the total B_tot (a), background B_ext (b), and local B_int (c) field maps. The local fields are equal to the total fields minus the background fields: B_int = B_tot - B_ext.

Figure 1: Offset dependence of MTsat and ihMTsat: Top: MTsat (--) increased strongly towards small offsets. Middle: The right flank of the ihMTsat histogram is assigned to WM. Its increases approximately parallel with WM MTsat. Bottom: The ihMT maps at ±2kHz (right) showed a higher level of noise than at 5kHz, as judged from the overlay and histogram width This is explained by the empirical signal equation (Eq. [1]) as high MTsat (likely enhanced by direct saturation) compromises noise propagation. The highest SNR in the ihMTsat map was generally obtained around ±5kHz.

Figure 3: Increasing the SNR of ihMTsat maps.

Top row: ihMTsat calculated from averaged gradient echoes reveals regional differences in WM.

Middle row: These are corroborated by averaging two acquisitions with a 20-channel head coil. The histogram reveals a distinct WM mode around 0.2 p.u..

Bottom row: SNR is further improved by a 32-channel head coil, ±4kHz frequency offset and 10° readout. A distinct WM mode at slightly higher ihMTsat is seen despite minor motion artifacts.

Figure 3: Influence of the FoV on $$$r_{\text{eff}}$$$. For six regions of the corpus callosum, box plots (horizontal lines in boxes denote median values; boxes include central 50 % of the values; whiskers include central 95 % of the values) of $$$r_{\text{eff}}$$$ for random lsLM subsections (black) are shown for varying FoV. The corresponding $$$r_{\text{eff}}$$$ of the whole lsLM sections (blue) and their consecutively cut mlEM subsections (red) are marked. The number of axons ranged from 1.0*10³ to 1.7*10³ in mlEM and 3.2*10⁵ to 1.3*10⁶ in lsLM.

Figure 4: Relationship between $$$r_{\text{eff}}$$$ and the fraction of large axons. Point markers denote estimations of $$$r_{\text{eff}}$$$ from large-scale light microscopy (lsLM) sections of a human corpus callosum. The linear regression line ($$$f_{\text{large}}$$$ [%] = -3.14 [%] + 2.85 * [1/µm] * $$$r_{\text{eff}}$$$ [µm], R² = 0.88) is plotted as a dashed line. The number of identified axons per region ranged from 2.5*10⁴ to 1.3*10⁶ (on average: 4*10⁵).

The hybrid model-derived intracellular water lifetimes from the simulated diffusion data with three SNR levels ($$$\infty$$$, 50, and 25) and two different imaging protocols, including $$$t_{diff}$$$ = 5, 30, 70, and 100 ms (blue circles) and $$$t_{diff}$$$ = 5, 30, 70, and 200 ms (black circles).

IMPULSED- and hybrid model-derived cell sizes from the simulated diffusion data with three SNR levels ( $$$\infty$$$, 50, and 25). The hybrid model was applied to data with two different imaging protocols, including $$$t_{diff}$$$ = 5, 30, 70, and 100 ms (blue circles) and $$$t_{diff}$$$ = 5, 30, 70, and 200 ms (black circles). The IMPULSED model was applied to data with $$$t_{diff}$$$ = 5, 30, 70, and 100 ms (green circles).

Effect of DFP on live MDA-MB-231 cell metabolism over time, as detected by ¹³C MRS. A: Average time course data of 1-13C-labeled Glc consumption and synthesis of [1-¹³C] Glyc, [1-13C] DHAP; [3-¹³C] G3P, [4-13C] Glu, [3-¹³C] Lac and [3-¹³C] Ala, are shown for DFP-treated (orange) and untreated (blue) MDA-MB-231 cells. B Average metabolite levels (mean ± SE) at different 6 h time intervals.

* Significantly different from untreated MDA-MB-231 cells (p<0.05, unpaired T test).

Comparison between the metabolite ¹³C-labeling rates of the human MDA-MB-231 and the murine 4T1 TNBC cells, (both, DFP-treated (DFP) and untreated (CTRL)), as detected by ¹³C MRS at each time point.

* Significantly different from untreated (CTRL) cells (p<0.05, unpaired T test).

^# Significantly different from DFP-treated cells (p<0.05, unpaired T test).

^¥ Significantly different from different time points (p<0.05, 2-way ANOVA with Geisser-Greenhouse correction

Figure 2: An example of matched ∆R2*(t) curves between in vivo and in silico data within a tumor and NAWM pixel (Fig 2a) and the corresponding anatomical T1-weighted and CBV maps (Fig 2b).

Table 1: Preliminary data demonstrating the ability to use the DRO as a benchmark for leakage correction algorithms. Using the consensus acquisition protocol, we computed CBV for an intact-BBB and a disrupted-BBB. We chose two leakage correction methods to compare for CBV calculation: Boxerman-Schmainda-Weisskoff (BSW) and the gamma-variate (GV) methods. Using our ground truth CBV maps, we can compute the concordance correlation coefficient (CCC) between various estimated CBVs across the 10,000 pixels.

Figure 1: Perfusion (a), water extraction fraction (b), permeability surface area product (c) and magnetization transfer ratio (d) maps from one control subject.

Figure 2: Bland-Altman plots for test-retest results for perfusion (a) and water extraction fraction (b). Correlation plots for test-retest results for perfusion (c) and extraction fraction (d).

Figure 1: Average CBF maps for all four scanner platforms. All maps are scaled the same which makes the baseline difference in global CBF calculation evident. The perfusion pattern however looks consistent across scanners.

Figure 2: Gray Matter CBF for each participant and site, highlighting scanner platform bias as well as covariation of CBF measurements across subjects despite the global bias.

Figure 3. ASL difference images of the slice #4 (a-b) and #5 (c-d) at flow rates of 175 (a) and 350 (b) mL/min. The slice #4 is at the second layers of porous material with 60 channels extended axially (off-center) in a circle. These 60 channels stop prior to slice #5. PCASL (1st column), VSASL using encoding directions of foot-head (FH, 2nd column), left-right (LR, 3rd direction) and oblique 45° (O45°, 4th column).

Figure 4. Ring ROIs and their normalized difference of slice #5 as a function of angular locations at flow rates of (a) 175 and (b) 350 mL/min, respectively. Normalized difference signal was averaged along the radius direction at every 0.5° from 0° to 360°. Half circle signal ranging from 45° to 225° was displayed for PCASL, VSASL with encoding directions of FH, LR and O45°. Green and yellow bars on the ring ROIs, corresponding to the green and yellow shades centered around 180° and 90° angles in VSASL (LR), indicate left and top sectors selected for labeling efficiency estimation in Figure 5.

Figure (1), In a) The arrangement of the neck coil alone around the neck of Duke model. In b) Configuration of the single system to be simulated.

Figure (2): Simulation data vs Experimental data. In (a) left, B1 simulation on a shperical phantom using the FDTD software, in (a) right, an actual B1 map from an MRI scanner after implementing the same shim. In (b) left, simulation of tuning and matching of a TTT panel, in (b) right, actual tuning and matching measurements of TTT panel on network analyzer

Figure 3 (A) The location of the labeling plane (red dashed line) shown on the Maximum Intensity Projection in the coronal view of the Time-of-Flight sequence. (B) The ROI was manually drawn on the Time-of-Flight image to cover the four inflow arteries (left and right Vertebral Arteries and left and right Internal Carotid Arteries). The simulated combined B1+ field of the (C) Bcg shimming and the (D) Maxmin shimming. The minimum B1+ amplitude within ROI was increased by 58% for Maxmin shimming. A dark band formed (yellow arrow), which would not affect the labeling efficiency.

Figure 4 Perfusion map acquired by pCASL sequences with labeling of Bcg shimming and Maxmin shimming. Increased perfusion was observed for Maxmin shimming. The top and bottom slices were discarded since they were cropped during preprocessing of motion correction.

Figure 2 Box and whisker plots depicting distribution of RBF (a) and split-RBF (b) measured with PC-MRI, ASL and DCE-MRI. The p-values for pairwise comparisons of means are given above the plots.

Figure 3 Bland-Altman plots comparing DCE-ASL (a), PC-DCE (b) and PC-ASL (c) for RBF (top panel) and split-RBF (bottom panel). Dashed lines indicate upper and lower 95% confidence intervals (CI) calculated as: mean difference ± 1.96 (SD). Solid lines represent the mean difference between two techniques.

Figure2. (A)(B) represent the result of conventional GRE and SE-WMRI respectively. In Ktrans maps, the cortex region has higher values. (Ktrans are 5544 and 5760 (1/min/1000) in conventional GRE and SE-WMRI.). In ve maps, the medulla area has high ve which is around 900 (1/1000) (Conventional GRE: 929, SE-WMRI:898 (1/1000)) but it is lower in the white arrow area. (C) shows the multi-region perfusion curves. The SE-WMRI curves are inconsistent because of physiological differences which also correspond to quantitative analysis.

Figure1. (A) demonstrates the three different stages of image result of conventional GRE. Same as (A), (B) show the image result of SE-WMRI. (C) presents the perfusion curves of conventional GRE and SE-WMRI. The image intensity is normalized to the pre-injection image intensity and apply the Gaussian low-pass filter to relieve the effect caused by motion. The blue line and red line represent the cortex and medulla region and the dots are the original data. The time delay of intensity peaks between the cortex and the medulla is 20.24s in conventional GRE, 30.72s in SE-WMRI.

Figure 5. Perfusion-Weighted Image and Signal of Kidney. A: Perfusion-weighted images of the kidney over different PLD times. B: Mean perfusion-weighted signal of the kidney cortex (red circle) and medulla (blue circle) over different PLD times.

Figure 3. Results of reconstructed images over time. A: Reconstructed images at the 10^th, 160^th, 310^th, 460^th, and 610^th frames. Yellow arrows indicate Moiré artifacts physically existing in the image due to the acquisition settings, which are not originating due to subspace-based reconstruction. B: Evolution of 1D profile at the center of liver (red line in A) over clock time (frame rate 34 ms, frames 301 to 600). Notice the change in liver position across time reflecting respiratory motion and the discontinuity in the liver position across time (yellow arrowhead) due to pCASL pulse.

Figure 2. The T2WI, EPI, and rCBV maps of a wild type (WT) and a knockout (KO) mouse overlaid with hippocampus ROIs (red) and whole brain ROIs (black).

Table 1. The mean rCBV in left, right, and total (left + right) hippocampus and the ratio of rCBV in left to right hippocampus in wild type (WT) and knockout (KO) mice.

Figure 1: Improved exchange time estimates. The proposed protocol provided improved fitting of exchange time parameter for brain regions with longer ATTs such as border zone areas.

Figure 3: Comparison of simulated exchange time standard deviation (SD) as a measure of estimate error. By concatenating datasets of two different sub-bolus duration, the resulting TIs provided lower exchange time estimate errors for later ATTs.

Fig.1 MRI analysis pipeline

Fig.2 Differences in perfusion parameter among NAWM, FLAIR and GD lesions

Table 1 Scanning parameters of GRASP sequence before and after optimization.

Table 2 Signal-to-noise ratio (SNR) of left lobe and right lobe of the liver before and after optimization in dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) of liver with GRASP sequence.

Figure 1: Preprocessing scheme

Figure 2: From volumetric measurements to gray matter clusters

Figure 1. The proposed CNN architecture for the estimation of pharmacokinetic parameters in the eTofts model. The input data are the 1D DCE-MRI time series data, $$$C_p$$$ (contrast agent concentration in the blood plasma) and $$$T_{10}$$$ ( $$$T_{1}$$$ before contrast agent injection) for each voxel. The number of filters and output nodes in the network are provided at the bottom of each layer. The outputs from the proposed CNN are the three independent eTofts parameters: $$$K^{trans}$$$, $$$v_p$$$ and $$$v_e$$$.

Figure 2. Example of the results from conventional eTofts model fitting using nonlinear-least-squares (NLLS) and CNN trained on mixed data in the testing dataset. (a) The contrast-enhanced slice of one glioma subject, where the tumor is outlined by solid red curves. (b) The predicted DCE-MRI time series signal using parameters obtained from conventional NLLS fitting (blue curve) and the CNN (red curve). (c) Pharmacokinetic parameter maps obtained from NLLS and the CNN, along with the absolute difference (error map) between the results from these two methods.

Figure 1: Exemplary case of a patient before (left column) and during anesthesia (right column)

Table 1: Main medication of the patients

Comparison of cerebral blood flow (CBF) between two time-points in the acupuncture group. Voxelweise analyses demonstrate a decrease in CBF in the right dorsal anterior cingulate cortex after treatment.

Comparison of cerebral blood flow (CBF) between two time-points in the sham acupuncture group. Voxelweise analyses demonstrate decreases in CBF in the left medial frontal gyrus, caudate, and insula after treatment.

Figure 1. CO₂ respiratory challenge with sinusoidally modulated stimulus in Subject 1. (A) Expected (red) and measured (blue) end-tidal CO₂ with 5mmHg amplitude in paradigm #1. (B) Expected and measured end-tidal CO₂ with 10mmHg amplitude in paradigm #2. (C) Whole brain BOLD time series in paradigm #1. (D) Whole brain BOLD time series in paradigm #2.

Figure 2. Perfusion maps derived from sinusoidal CO₂ challenge in Subject 1. CBV, TT and CBF maps in paradigm #1 (top row) and paradigm #2 (bottom row).

FIG. 4 - (A) Areas showing an increase in resting CBF with learning reported as p-value. (B) Mean±SEM resting CBF during the different resting periods (pre-/post- task/control) in significant voxels. Resting CBF significantly increased following completion of the task 15.8% on average compared to the pre-task.

FIG. 1 - Participants were scanned twice. During the “task” session (A), participants underwent 8 minutes of resting state (RS), 10 minutes of motor task and 8 minutes of RS. During the “control” session (B), participants underwent 8 minutes of RS, 10 minutes where they were asked to lay still in the scanner and 8 minutes of RS again. During the rest scans, in the task session and for the entire duration of the control session, a white fixation cross and the word “REST” were presented on a black screen.

Figure 1: Box plots showing the distribution of: A) the subject-specific T₁ values with nominal T₁ values for blood (1.65 s) and tissue (1.33 s) plotted in brown and blue respectively; B) cerebral blood flow (CBF) calculated with nominal and subject-specific T₁ values, mean CBF values using the estimated T₁ values are plotted in green (grey matter, GM) and purple (white matter, WM).

Figure 2: Bland-Altman plots of cerebral blood flow (CBF) values in grey (GM) and white matter (WM) calculated using the calculated using standard (tissue=1.33 s, blood=1.65 s) and subject-specific T₁ values.