Figure 1: Schema of transport and degradation of D-Glc, L-Glc, 3OMG and 2DG
After injection into the vascular space, molecules of interest (except L-Glc) are transported into the extravascular/extracellular space, then into the intracellular space by GLUTs transporters. D-Glc is transported in cells where it can be metabolized through glycolysis, penthose phosphate pathway (PPP) or into glycogen. 3OMG is transported in cells but does not undergo further metabolism. 2DG is transported in cells where it metabolized into 2DG6P by hexokinase but it does not undergo glycolysis.

Figure 3: in vivo kinetics of glucoCEST signal
Intravenous injection of D-Glc (A), L-Glc (B), 3OMG (C) and 2DG (D). GlucoCEST variations were measured before and after intravenous injection (gray area) at 0.8 ppm, 1.2 ppm, 2.1 ppm, 2.9 ppm. Mean ± SEM variations of CEST signal compared to baseline are shown (n=5, 3, 9 and 7 respectively).

Figure 2: Representative motion pattern (black) and estimates of conventional image registration (red), RPCA+PCA_R (blue) and proposed method (green) with the five offsets close to the water resonance (i.e. ±0.5 ppm) highlighted (A). Corresponding image misalignment (B) and spectral error (C) with and without motion correction (colors and black respectively).

Figure 3: Statistical analysis for n=100 different motion patterns: boxplots of the overall spectral error (A) and the maximum misalignment (B) for uncorrected (black) and motion corrected data using the conventional image registration (red), RPCA+PCA_R (blue) or proposed method (green).

Figure 1. Schematic illustration of diffusion-weighted CEST. 3D images were acquired at 6 offsets around 3.5ppm and -3.5ppm (6 point acquisition¹) and at 11 b-values = 0, 5, … 45, 50 s/mm².

Figure 5. Diffusion maps of human brain. Axial anatomical images in three slices (each row) are shown (a, b, c). MTR asymmetry maps at 3.5 ppm demonstrates CEST distribution in human brain (d, e, f). Apparent diffusion coefficient map was calculated and the average value was 0.001mm²/s across the slices (g, h, i). White matter tracks were discernable in the CEST-weighted apparent diffusion coefficient map (j, k, l).

Fig. 2 The architecture of DL-B0GluCEST-HS is an enhanced deep residual network. The first layer consists of 32 convolutional filters with 3×3 kernel size for each input image. After concatenating them as one channel and going through another convolutional layer, the subsequent layers include 8 consecutive WDSR blocks, which contains 2 convolutional layers and 1 activation layer. Another convolutional layer was attached to the end to get the B0 corrected ±3 ppm image with additional input from the concatenated layer. The subtraction is calculated to obtain B0 corrected CEST image.

Fig. 4 GluCEST ratio maps of a subject were calculated by different methods. Row (a)(c)(e) are GluCEST results and row (b)(d)(f) are differences map between the labeled method and the gold standard conventional approach [10]. The number in the methods indicates how many downfield offset images were used as the input to the DL networks.

Figure 2. Multipool Lorentzian fitting of (a, b) apparent and (c, d) QUASS Z-spectra in (a, c) contralateral normal tissue and (b, d) tumor regions with Ts/Td of 2s/2s and 4s/4s, respectively.

Figure 3. Multi-parametric MR images of a representative rat bearing C6 glioma tumor. (a) T1, (f) R1, apparent CEST maps of (b, c) MT and (d, e) APT, and QUASS CEST maps of (g, h) MT and (i, j) APT with Ts/Td of 2s/2s and 4s/4s, respectively.

Figure 4. GlucoCESL parameter maps from a tumour bearing rat using model 1 (a-d) and model 2 (i-m), and mean parameter values in cortex (healthy – triangle; tumour bearing – circle; n = 7) and tumour (n = 4) for model 1 (e-h) and model 2 (n-r). Far right is the corresponding ΔAIC map (s) and mean ΔAIC in cortex and tumour (t). T-tests for partially overlapping samples were used to test the null hypothesis of no difference in parameter values between cortex and tumour.

Figure 3. a) An example CESL image acquired with TSL = 0. These images are T2-weighted and provide excellent contrast for definition of tumour (yellow) and cortex (green) ROIs which were drawn manually in MRIcron. Example voxelwise data and kinetic model fits in the cortex (b) and tumour (c). Example ROI averaged data and fits in the cortex (d) and tumour (e). Both models provide a good fit to cortex data. Model 2 appears to provide a better fit to tumour data.

Figure 1: Coronal and Axial views of MTR and ihMTR as a function of pulse B1+RMS. Shaded regions indicate ROIs used for MTR and ihMTR measurements in Table 2. Larger % ihMTR is visible in the denser white matter tracts, as compared to MTR which is relatively more homogeneous across all white matter.

Table 2: Observed and simulated MTR and ihMTR versus applied B1+RMS in two sample ROIs.

Figure 5: Dictionary fits to in vivo data with and without motion correction. Motion artefacts are significantly reduced for the former, enabling quantification of T_1D^s. The lower contrast-to-noise ratio of ihMT compared to MT means quantification of dipolar parameters is hindered prior to correction. GM-WM contrast appears enhanced for T_1D^s versus f and highly myelinated structures become more discernible. Semisolid relaxation times were fixed at T_1Z^s = 0.2s and T₂^s = 7.5μs; free pool T₂ at T₂^f = 69ms; exchange rate at K = 50s^-1 and main magnetic field induced phase at ΔB₀ = 0^13,14.

Figure 4: Top: Estimated motion traces during the in vivo acquisition. Results are from a compliant volunteer who was instructed not to move during the acquisition. "Tra" refers to transversal motion and "Rot" refers to rotation; "LR" indicates the left-right direction, "AP" is anterior-posterior, and "FH" is foot-head. Bottom: Example in vivo contrast ratio maps with and without motion correction from a central axial slice and reconstructed using the first eight singular components. MTR shows strong grey matter (GM)-white matter (WM) contrast and ihMTR is more correlated to WM.

Figure 5) Results of the in vivo measurements on mice (short axis view, isotropic resolution 250μm). In a) maps are shown using monoexponential fitting (left) and corrected fitting (right) for T_rec=1275±66ms. The correlation with the different recovery times are shown in b). Here we considered regions of interest in myocardial tissue (left) and hepatic tissue (right). Positive correlation can be observed for monoexponential fitting, while the corrected fit provides a reduced correlation. The T₁ in myocardial tissue was 1388±51ms. (t_SL=4,12,20,28,36,44,52,60ms, f_SL=1500Hz)

Figure 4) Results of the phantom measurements. In a) and b) the results of T_1ρ mapping (BSA 15%) are shown using Eq.1 (monoexponential) and Eq.2 (corrected) respectively. In c) the results are shown for all BSA concentrations. Monoexponential fitting leads to an underestimation of T_1ρ for small recovery times. The corrected fit using the corresponding T₁ values and recovery times prevents the underestimation. The mean quantification error could be reduced from -7.4% to -1.3%. (t_SL=4,17,30,43,56,69,82,95ms, f_SL=1500Hz)

Figure 1: Quantitative T2, cwT1ρ, αT1ρ, and αT2ρ maps for one of the piglets. Also shown is subtracted contrast-enhanced MRI, which shows a clear lack of perfusion to the ischemic femoral head (asterisk), and the region of interest masks used to quantify the regional relaxation time values. T2, cwT1ρ, and αT2ρ are all noticeably increased in the secondary ossification center (SOC) of the ischemic vs. contralateral-control femoral head. All four relaxation times are increased in the epiphyseal cartilage.

Table 2: Region of interest measurements for five pairs of ischemic and control femoral heads. Values shown as mean ± standard deviation. * p<0.05.

Figure 2: DeepCEST network trained on healthy volunteers applied on a healty volunteer (a) Comparison of the predictions and the Lorentzian fit amplitudes (ground truth). (b) Error of the prediction compared to the fit. (c) Uncertainty of the prediction.

Figure 1: a) Input data: uncorrected Z-spectra of two B₁ levels acquired at 3T. b) Target data: 5-pool-Lorentzian fitted data acquired at 7T. As targets only the Lorentzian amplitudes were used.

Figure 2: Comparison of retrospective and prospective input feature reduction as well as conventional Lorentzian fitting. (A) left to right: ground truth by Lorentzian fitting on full measurement, linear prediction with all input features contained, linear prediction with retrospective reduction of input features according to the LASSO-reduced sampling schedule, measurement with these input features being prospectively omitted. (B) Difference maps of the linear predictions and the ground truth. (C) B1-MIMOSA and (D) B1-cp map as mentioned in ⁵

Figure 3: Comparison of retrospective reduction of input features for a tumor patient dataset (Glioblastoma WHO grade 4). (A) left to right: ground truth obtained by Lorentzian fitting, linear prediction with all input features retained, retrospective reduction of input features. (B) Difference maps between linear predictions and ground truth. (C) B1 MIMOSA map and (D) B1-cp map as mentioned in ⁵. (E) Anatomical T1 contrast-enhanced image

Figure 2: Data measured (spatial mean and SD) at B₀ = 14T and simulations (same pulseq-file). CEST pools included water (T₁=1735ms, T₂=18ms, f=100%), ssMT (T₁=1000ms, T₂=0.01ms, dω=0ppm, k_ex=23Hz, f=5%), L-arginine (T₁=1000ms, T₂=5ms, dω=3ppm, f=2.25%) and agar-agar (10) (T₁=1000ms, T₂=50ms, dω=1.6ppm, k_ex=6500Hz, f=1%). Details on pre-saturation and simulation source code: https://pulseq-cest.github.io/ file “APTw_3T_002”; modifications: dω list and B₀. Larger deviations in MTR assymetry at <= 1.5 ppm are most likely due to residual B₀ variation in the measured data.

Figure 4: (left) ROI averaged (mean ± SD) Z-spectra acquired for different CEST pres-aturation modules using Bruker’s FLASH readout. (right) MTR asymmetry.

Figure 4. Results of linear projection in a tumor patient test dataset. Clinical contrasts: (A) T1 weighted contrast-enhanced, (B) MPRAGE and (C) FLAIR. (D) Ground truth Lorentzian fit results. (E) Contrast maps obtained by linear projection with coefficients obtained from 5 healthy subject datasets. (F) Difference maps to ground truth for linear projection result. (G) Voxel-wise scatter plots of linear prediction result (y) versus ground truth (x) with legends indicating Pearson correlation coefficient r.

Figure 1. Analogy of discrete Fourier transform and the proposed linear projection approach for CEST evaluation. For the Fourier transform, a signal vector S(t) is projected onto basis vectors $$$\beta_1,...,\beta_n$$$ consisting of the respective harmonics to yield the Fourier coefficients $$$A_1,...,A_n$$$ for different frequencies. For linear CEST evaluation, acquired raw data are projected onto coefficient vectors found by linear regression applied to conventionally evaluated data, to yield target contrasts like APT, NOE and ssMT amplitudes.

Fig. 1 AutoCEST pipeline. a. Pre-experiment step. The broad expected range of properties (blue rectangles) are given as input to an MR physics governed AI system, which dynamically optimizes the protocol parameters (orange rectangles). The resulting “ADC” MR signals are then decoded into quantitative parameters using a deep reconstruction network. b. Experiment step. The optimal schedule parameters are loaded into the scanner, resulting in a set of N raw images. The resulting images are fed voxelwise into the trained reconstruction network, resulting in quantitative CEST maps.

Fig 5 In vivo study results. (Left). T₂-weighted image of a tumor-bearing mouse. (Center). The corresponding rNOE volume fraction quantitative map (f_s), and its overlay (right) atop the tumor and contralateral ROIs. Note the decreased f_s at the tumor.

Figure 2: Representative NOE, APT and MT maps from Multitasking ss-CEST. Three different slices are displayed to show the consistency of image quality within the 3D volume.

Figure 4: Average Lorentzian amplitudes within GM and WM regions among different volunteers. The mean amplitude is consistent among healthy subjects. Contrast ratios of WM:GM for NOE/APT/MT: 1.21/1.10/1.22 (Multitasking ss-CEST) vs. 1.20/1.02/1.35 (2D single-shot FLASH CEST).

Figure 1. The reference images using standard recon (a) and DL Recon (d). The MTR_asym map at 3.5 ppm (b,e) and averaged between 3.0-4.0 ppm (c,f) using standard recon (b,c) and DL recon (e,f) of a glioma patient. The tumor ROI (black) and the background ROI of homogenous brain tissue (red) was shown in (a) as an example.

Figure 3. The tumor ROI-averaged MTR_asym divided by the standard deviation of the background ROI at 3.5 ppm (a) and 3.0-4.0 ppm (b) using standard recon and DL Recon were compared for all patients. (a) p = 0.002. (b) p = 0.016. **: p < 0.01. *: p < 0.05.

Fig. 1: ΔCrCEST time-series for volunteer #1, corresponding to the middle slice (5th out of 8 slices). Each row corresponds to different exercise workload. Selected time frames of post-exercise CrCEST time-series are being presented, with frame number written on right top corner of each frame. The temporal resolution was 30s. Though at mild PFE level, the volunteer mostly utilized the lateral gastrocnemius (LG), both LG and medial gastrocnemius (MG) were utilized during intense PFE. As evident visually, the recovery rate increased with increased intensity of exercise levels.

Fig. 2: Panel -1: Time-series of Lactate CEST (LATEST) maps show intense exercise-induced changes in skeletal muscle, using images from 5th slice of volunteer #1. First LATEST frame corresponds to pre-exercise level, whereas the remaining ten frames correspond to ten time points post-exercise with 140s temporal resolution. Panel-2: Time-series plots of LATEST values averaged over the entire lateral gastrocnemius and corresponding fits are shown.

Parameter maps and uncertainty estimation of simulated phantom data. The left column shows the reference maps with simulated B₀- and B₁- inhomogeneities and different brain-matter-like T₁ values with 1% Gaussian noise overall and one compartment with noise ranging from 0% to 10%. From second left to right: maps generated by the NN, difference maps between reference and NN output (x10) and uncertainty maps generated by the NN (x10).

Exchange-weighted contrasts of simulated CEST data before and after post-processing. Simulated with B₀- and B₁- inhomogeneities, different brain-matter-like compartments and a three-pool-system with the same CEST-pool fraction of f = 0.6% overall and one compartment with fractions ranging from 0% to 1%. Left to right: MTR_Rex⁷ contrasts of i) the uncorrected CEST-simulation and after ii) ΔB₀-correction⁵, iii) B₁-correction⁶, iv) ΔB₀- and B₁-correction. Right: AREX contrast (ΔB₀-, B₁- and T₁-corrected)⁷.

Fig. 2., Simulation of the proposed QUASS CEST MRI. a) R1rho determined from the QUASS solution for three representative sets of Td and Ts (i.e., 1s/1s (cross markers), 2s/2s (diamond markers), and 7.5s/7.5s (plus markers) and that calculated under long Td and Ts (7.5s/7.5s, black plus markers) for B1 levels of 1 and 2 µT. b) The QUASS Z-spectra under the three representative sets of Td and Ts and two B1 levels of 1 and 2 µT. c) The QUASS CEST effect at 1.9 ppm as a function of B1 level for three sets of Ts and Td times, being 1s/1s, 2s/2s, and 7.5s/7.5s.

Fig. 4., The relationship between Td/Ts with the experimentally determined ksw and fr. a) ksw from the conventional CEST MRI (circle markers) showed a significant linear dependence with Ts/Td. In comparison, ksw determined from the QUASS CEST MRI (square markers) did not show significant dependence with respect to Ts/Td. c) fr determined from conventional apparent CEST MRI (circle markers) can be described by a stretched exponential function with significant dependence on Ts/Td. d) fr determined from the QUASS CEST MRI (square markers) showed no significant dependence on Ts/Td.

Figure 4. A typical example of the effect of MTC on the VDMP build-up curves from the human WM for B₁ = 1.99µT. In A) the saturation build-up of the different CEST-pools is plotted together with the MT contrast. In B) the MTC has been removed and the trend of the VDMP build-up curves from the CEST-pools with different exchange rates can be distinguished.

Figure 2. Normalized VDMP saturation build-up curves for different mixing times. A) Simulated build-up curves for CEST-pools found in the human brain: according to the concentration of amines in the GM (dotted line) and WM (dashed line) B) Build-up curves from glutamate (amines in blue) and bovine serum albumin phantoms (amides in yellow and rNOE in green). Both simulation and phantom data show a gradual build-up for slow exchanging molecules (rNOE and amides) as opposed to a fast decay from fast exchanging amines.

Fig3 (a) CEST labeling image at 3.5ppm, (b) Z-spectrum with marked as blue circle in (a), and (c) MTR asymmetric map at 3.5ppm.

Fig2 (a) Reconstructed (unsaturated) images acquired from the proposed acquisition, (b) Saturated Z-spectra images offset frequencies -10ppm to 10ppm

Figure 3: In vivo application of the calibrated pH-CEST technique in tumor-bearing mice. pH values of 6.97 ± 0.09, 7.00 ± 0.20 and 7.13 ± 0.19 were found in the subcutaneous lesions of three DLD xenografted nude mice.

Figure 2: Correlation of calculated pH from CEST-MRI with titrated pH (A,C) and tissue concentration (B,D) of porcine brain lysate. C: Correlation of the titrated pH and the mean pH values calculated from the CEST_ratio. D: Mean pH values calculated from the CEST_ratio as a function of concentration.

a) APT maps and b) NOE maps of a repeated measurement (group 1) before and after improving the post-processing steps (colorbar adjusted for visual comparison) c) decreasing CoV as a function of the stepwise improvement of post-processing methods: 1) moco with interpolation, 2) additionally, 2 point M₀ normalization, 3) moco to the offset at 3.5 ppm, 4) B₀ correction by using a Lorentzian fit 5) B₀ correction with spline interpolation parameter 0.999 6) 2 point M_Z normalization 7) LFM 4

Conventional images (contrast-enhanced T₁ weighted imaging, MPRage, and Flair) and CEST APT image of an epilepsy-associated tumor initially diagnosed as low grade, histologically proven to be glioblastoma, IDH wild-type. Improved CEST postprocessing pipeline was used.

RF magnitude A) and phase B) of the spin-lock saturation at 0.6 ppm. The frequency modulation can be seen from the changing phase in the zoomed plot (black rectangle). The red dot marks the beginning of the readout period. RF magnitude (C,E,G) and phase (D,F,H) during three different APTw protocols: APTw_3T_001 (C,D), APTw_3T_002 (E,F) and APTw_3T_003 (G, H) all with B_1,cwpe = 2 µT and recovery time, T_rec = 3.5 s. In C) and D), a zoomed graph for two RF pulses is shown in the black rectangles. The phase accumulation due to the off-resonance of the RF pulses is taken into account (D).

MTR_asym maps at 3.5 ppm for the APTw_3T_001 (A), APTw_3T_002 (B) and APTw_3T_003 (C) protocols. Protocols were run with repeated acquisitions (3 repetitions) at the offsets of interest and a dummy scan at the beginning.

Figure 3. (a-c) Values of CRB-, DP-, ED-based indexes with varied acquisition number of 2-30; (d) RMSE of pH in MC simulation study using both DP- and ED-based matching methods; RMSE of (e) log(kex) and (f) respective pH values of the phantoms; the CEST-MRF measured pH maps in the phantom study using both (g-i) DP- and (j-l) ED-based matching methods for representative acquisition numbers of (g, j) 6, (h, k) 18 and (i, l) 30.

Figure 1. (a-c) Alterations of CRB-, DP-, ED-based metrics with exchange rates; RMSE of pH quantification in MC simulation study using both DP- and ED-based matching methods under noise levels of (d )20, (e) 40 and (f) 55 dB; RMSE of (g) log(kex) and (h) pH of the phantom; (i) correlation of the measured and the titrated pH values for vials with the same concentration of 60mM.

Fig. 2. Representative examples of structural image and CEST maps of one mouse brain. (a)T₂W image with tumor region delineated by the red dotted line. (b), (c) Representative proton exchange rate and labile proton ratio maps. (d), (e), (f) pH weighted contrasts calculated using pH enhanced method, AACID method and amide proton transfer (APT) MRI.

Fig. 3. Association between labile proton ratio (f_b) , chemical exchange rate (k_b) values and AACID (a) (d) ,pH_enh (b) (e), APT(c) (f) values of all voxels in 6 mice brain from tumors (red points) and normal tissue (blue points).

Figure 3: Typical multi-slice human brain images acquired with starCEST at 3T. (a) T₁ maps by look-locker sequence; (b) B₀ maps acquired with dual-echo sequence, The (c) amideCEST maps (ΔZ_amide) and the corresponding (d) apparent relaxation rate (R_amide) maps extracted with the PLOF method by including the B₀ and T₁ maps, respectively.

Comparison of the amideCEST maps without and with MLSVD post-processing. (a) The original amide CEST maps extracted using PLOF without MLSVD denoising. (b) MLSVD denoising applied with truncation numbers 48, 48 and 10.

Figure 1: Representative segmentation masks, T₂-weighted images and histology (H+E and TUNEL stains) for each category of response to radiation treatment (responder, partial responder, non-responder).

Figure 4: Average estimated qMT parameters by treatment response of (a-d) active and (e-h) necrotic/apoptotic regions. Free parameters were: T₁ and T₂ of the water pool (T_1,L and T_2,L respectively), the exchange rate of magnetization from the MT pool to the water pool (R_MT), original magnetization of the MT pool relative to the water pool (M_0,MT), and T₂ of the MT pool (T_2,MT). *p < 0.05. **p < 0.01.

Figure 1: (Top row) CrCEST time course imaging (left) ,and ASL (middle), and overall CrCEST decay (right) in a patient with an ABI of 0.68 prior to endovascular revascularization.(Bottom row) CrCEST time course imaging (left) , ASL (middle), and overall CrCEST decay (right) CrCEST time course imaging and ASL in the same patient 6 months after revascularization, showing changes in both metabolism and perfusion.

Table 1: CEST and ASL evaluation pre and post revascularization in patient 1. ASL and CEST measurements are reported as a mean of the muscle ROI in which the greatest perfusion was seen in the ASL images.

Figure 1. The flow diagram of the proposed method PLANT.

The T1ρ-weighted images at TSL=1ms reconstructed using PLANT, HD-PROST and L+S with different acceleration factors (R=4, 5 and 6) from retrospectively undersampled data. The corresponding error maps are also shown for comparison. The reference image is obtained from the fully sampled k-space data.

Figure 3. a) A raw image acquired using AC-iTIP at TSL 0 ms. b) The fitting results from the selected ROI in a). c) The ground-truth T1rho map acquired. d) and e) show the T1rho map before and after correction, respectively. f) compares the T1rho values in c)-e). Blue, red and yellows bars represent c), d) and e), respectively.

Figure 4. a)-c) and d)-e) show B1 and B0 field maps from three different scans. g)-i) and j)-l) show the T1rho maps in vivo before and after correction. The cartilage and muscle are manually drawn for analysis.

Fig 1. a) Schematic diagram of an imaging voxel illustrating the chemical exchange of labile hydrogen, between water and hydroxyls, occurring in the presence of susceptibility gradients generated by the red blood cells containing paramagnetic deoxyhemoglobin. b) Simulated R_1ρ dispersions for a spherical structure with a radius of 10 μm. The low-frequency dispersion corresponds to diffusion effects (blue), the high frequency corresponds to exchange (red), and the composite curve “double dispersion” added two rates linearly R_1ρ = Rdipolar + R_1ρ^Diff + R_1ρ^Ex (adapted from Ref 3).

Fig 5. 3D T_1ρ images collected with the anterior coil, Slice N=3, TR = 1250 ms, shortest TE, SENSE factor 2, a FOV of 224 x 224 mm, and a voxel size of 1 x1 x 8 mm with TSL; 2, 12, 36 ms and FSL=0, 60, 120, 200, 300, 480 Hz resulted in a total scan time of 7 min and 56 sec. (a) Example of anatomy and R_1ρ=1/T_1ρ maps from slice #3. ROI placed in the area that has approximately B₀ off-set less than 30Hz. (b) Averaged R_1ρ values from selected ROI. The error bars show the standard deviation. The fitted curve calculated using R_1ρ equation (c) Fitted parameters extracted from R_1ρ dispersion depicted in (b).

Figure 4 Sample images of reference and accelerated T_1ρ map. The T_1ρ map of cartilage compartments was overlaid to the corresponding DESS images for better visualization.

Figure 3 (a) shows the CV between reference and accelerated T_1ρ value, scan-rescan CV, and scan-rescan ICC with a 95% confidence interval. (b) shows the Bland-Altman plot between accelerated and reference T_1ρ value. Each entry corresponds to an average value of a cartilage compartment in a subject. ICC was calculated for each AF.

Figure 3

Representative T_1ρ-prepared images (only SL-time=30 ms shown in top row) and T_1ρ map (bottom row) obtained from the same acquisition. Homogenous T_1ρ contrast can be observed in the whole brain, even in lower structures for SL-time=30 ms. Occasionally some artifacts may be observed above the nasal cavity due to air-tissue boundaries (arrow). Quantification of T_1ρ also shows good contrast and homogeneity throughout the brain.

Figure 4

Example sagittal slices from T_1ρ-prepared volumes and the resulting map in the same subject during Reference, Repeat and One-Week scan show increased T_1ρ weighting as function of SL-time as well as good visual repeatability and reproducibility of image quality between scans.

FIG.3 Maps of brain (a), B₀ (b), B₁ (c), and ΔR_1ρ (the subtraction of two R_1ρ maps acquired with ω₁ of 0 and 100Hz), without/with correction (f, g) from a healthy human subject. (d, e) show the averaged R_1ρ dispersions without/with correction from the ROI with large B₀ shift (│Δω_off│ > 20 Hz) (h) and small B₀ shift (│Δω_off│ < 20 Hz) (i), respectively.

FIG. 2. Single-pool (a, c, e) and two-pool (b, d, f) model simulated R_1ρ dispersions under B₀ shift (a, b), B₁ shift (c, d), and both B₀ and B₁ shift (e, f), and with correction (green lines) and without correction (red lines). R_1ρ dispersions with no shifts in either B₀ or B₁ (blue lines) were also plotted for comparison. Simulation parameters included T_1a=1.5s, T_2a=80ms, T_1b=1s, T_2b=30ms; fractional concentration of pool B is 0.2; the exchange rate from pool B to pool A is 20s^-1; resonance frequency of pool A is 0 Hz and pool B is 63.5 Hz.

Figure 1. (a). CW-T1ρ preparation is implemented by a composite RF pulse (90°-TSL/2-180°-TSL/2-90°). Phase cycling of CW-T1ρ is achieved by a phase reversal of the last 90° pulse. (b). Adb-T1ρ is achieved by a series of four adiabatic full passage (AFP) pulse group following MLEV phase cycling. The frequency modulation of the second half of the last AFP pulse is reversed for phase cycling. N: number of repeats of the four AFP pulse group; VFA: variable flip angle; TSL: time of spin lock, Tsr: time for saturation recovery. AM: amplitude modulation; FM: frequency modulation.

Figure 2. Example phantom T1ρ maps from 3D MAPSS CW-T1ρ (left) and Adb-T1ρ (right) at the center slice 24 (top row), off-center slice 8 (middle row) and off-center slice 41 (bottom row). Slice 8 and slice 41 of CW-T1ρ clearly show non-uniform T1ρ maps due to artifacts from frequency offset, while the T1ρ maps from Adb-T1ρ are more uniform and consistent at the two off-center slices. CW-T1ρ: continuous-wave T1ρ mapping; Adb-T1ρ: adiabatic T1ρ mapping.

Figure 2. Representative T_1ρ maps (first row) and corresponding T₂w-FSE images (second row) of four subjects (as shown in four columns respectively). These T_1ρ maps demonstrate that UTE-Adiab-T_1ρ sequence allows measurement of T_1ρs of the whole IVD, including the CEP. In comparison, the CEP regions were low signal on the T₂w-FSE images. The modified Pfirrmann grades of discs are included on the T₂w-FSE images.

Figure 3. Spearman’s correlation coefficient results between modified Pfirrmann grades and T_1ρ values of OPAF (a), SCEP (b), ICEP (c), and NP (d). A strong negative correlation is observed in the NP, and moderate positive correlations are observed in the OPAF, SCEP, and ICEP.

Figure 4 IhMT images and maps of selected qihMT parameters output from global fits to ihMT data acquired with optimized protocol in central nervous system tissue from donors with and without MS. The ihMT images in the top row are formed from the difference between the summation of data prepared with single or dual frequency RF irradiation applied off-resonance. Color bars on the right relate to the maps of qihMT parameters in the same row all with lowest levels of 0 and highest indicated. All images and maps were resized to twice their resolution by nearest-neighbor interpolation.

Figure 1 Schematic of MT sequence (top) for ihMT experiments illustrating some of the scan parameters open to optimization and table of optimal protocol of scan parameters (bottom) from CRLB analysis. Optimized parameters included: MT preparation duration (𝛕_RF), recovery time (𝛕_rec), root mean square B₁ (B_1,RMS) over the MT preparation, and frequency (𝛥), pulse width (pw), and duty cycle (DC) of MT pulses.

Figure 2. Representative maps of: a,c) ihMTR; b,d) ihMTR_inv; a,b) maps calculated by 3D segmented SPGR dataset and c,d) maps calculated by 3D segmented SPGR-EPI dataset from a healthy volunteer, as acquired in the sagittal plane, and reformatted into the coronal and axial planes.

Figure 1. Sequence diagrams used to acquire data for: a) ihMT, and b) reference using a T1 preparation.

Fig 2. Example M₀^b outputs. The first column shows the output of the non-linear least-squares fit of the Bloch McConnell equations used as the ground truth. Subsequent columns show the ANN, CNN, and Hybrid model outputs. The second row contains the difference maps of the associated model and Bloch-McConnell fit.

Fig 3. Model performance. A) Example segmentation of GTV and cNAWM. B) and C) show Bland Altman plots of the single slice in A, and the full test set, respectively. Scatterplots show the relationship between network predictions and the non-linear least-squares fit of the Bloch-McConnell equations, and the Pearson correlation coefficient for each ROI. The difference plots evaluate the bias (solid line, dashed lines 95% CI) between mean differences of the network and ground truth in each ROI.

Figure 4. MT studies of CA:SD:BTAC. MT is generated by exchange of water protons with amide protons of SD and hydroxyl protons of CA. MTA is relatively small as the two exchange mechanisms balance. Note that the red line of the figure, (+-) RF saturation, is not visible as the yellow line, (-+) RF saturation, is identical and overwrites the (+-) line. T1d in this sample is long and eMT is much larger that conventional MT. This leads to a large ihMT signal.

Figure 6. MT studies of CA:SD:BTAC plus cholesterol. Cholesterol provides an additional hydroxyl proton for cross-relaxation and a different underlying chemical shift profile. Addition of rigid cholesterol increases MTA relative to pure CA:SD:BTAC shown in Fig. 4. The MT asymmetry of cholesterol is also very broad. ihMTR is decreased as cholesterol stiffens the membrane and decreases proton T1d times, allowing for more efficient intermolecular spin diffusion in the lipid matrix.

Figure 2: Spectra acquired at 7T at 5 different powers and fitted using the PSO algorithm, with an acquisition time of 9s/frequency.

Figure 1: Spectra acquired at 3T for the diverse concentration and at 3 different saturation powers and fitted using the PSO algorithm, with an acquisition time of 21s per frequency.

Figure 1. A 65-year-old male patient with lymph node metastasis of rectal cancer. T2W image (a), DWI (b), T1 map (c), and APTw-T2W fusion image (d) to show the lesion of rectal cancer with lymph node metastasis. Three ROI of rectal cancer were showed on T2W image. APT and T1 values of the ROI were 0.89%, 0.45%, 0.97% and 1579.90ms, 1449.80ms, 1496.05ms.

Figure 2. A 64-year-old male patient without lymph node metastasis of rectal cancer. T2W image (a), DWI (b), T1 map (c), and APTw-T2W fusion image (d) to show the lesion of rectal cancer without lymph node metastasis. Three ROI of rectal cancer were showed on T2W image. APT and T1 values of the ROI were 2.03%, 2.64%, 1.73% and 1371.23ms, 1337.87ms, 1353.56ms.

Figure2: (A) The illustration of B₁ and RF offsets sampling and correction methods for conventional, 4 points, B0B1CAR, and full correction methods, respective. (B) The sagittal APTw images with B₀ and B₁ correction by the corresponding methods. The mean errors in the upper brain region (UB error) and the whole brain region (WB error) are measured by the mean difference[LP(DMRS1] with the full correction method. The relative total acquisition times (TA) are compared with the conventional method.

Figure1: (A) Conventional method for CEST B₀ and B₁ correction. First, the data at the targeted z-spectrum frequency was interpolated at each B₁ value. Then, the data at the targeted frequency and B₁ were interpolated from the B₀ corrected data. (B) B0B1COR method for CEST B₀ and B₁ correction. The data at the targeted z-spectrum frequency and B₁ was interpolated directly from nearby sampled z-spectrum at different B₁ values.

Figure 1 – Diagram of the proposed radial pulse sequence. For each time step, a saturation RF pulse train is played. The resonance frequency of the pulse train can be fixed or varied according to the experiment. Each pulse train is then followed by a radial acquisition of 16 spokes whose angle is defined by the golden-angle.

Figure 4 – CEST-MRF acquisition with the proposed sequence. Note the variation in signal intensity due to the varying saturation pulse power.

Figure 1. Simulation shows MTR asymmetry concentration dependence at 1.9 ppm 10 % duty cycle π pulses with avg. B₁ = 0.48 μT (a). MTR asymmetry measured in creatine phantoms at varying pH using different duty cycles with avg. B₁ = 0.48 μT (b) show increasing pH dependence with increasing duty cycle. In creatine phantoms with varying concentration and varied pH, the average R_ex asymmetry with an avg. B₁ = 0.97 μT using continuous wave saturation showed relatively weak linear dependence on concentration (c) while a 10% DC π pulse showed strong linear dependence on concentration (d).

Figure 2: Histogram of CEST contrast at 3ppm offset at B_1rms of 4.2μT and 2s pulse duration, from 10mM Glutamate and 10mM Glutamine phantom at pH 7 is shown in blue color while the one scaled to in vivo physiological concentration is shown in green color.

Figure 1: Molecular structures of Glutamate and Glutamine with the primary side chain differences, carboxylate vs amide circled in red color.

Tabel 1: The comparison of several motion-correction methods for CEST images with simulated motion. No_reg: without registration ; Reg2S0: registration to mean image; Reg2mean: registration to mean image; Regto3.5ppm: registration to the 3.5ppm image; Regto-3.5ppm: registration to the -3.5ppm image; Reg2lowRank: iterative registration to low-rank approximated image.

Figure 1: The S0 image, Z-spectrum of the rat03 tumor region without simulated motion and Z-spectrum without motion correction were shown at the first line. The spectra that were corrected for motion from the CEST images with simulated motion by different methods were displayed at the rest lines.

Figure 2. Amide-proton-transfer-weighted (APTw) images obtained from the bovine serum albumin (BSA) phantom with the maximum achievable saturation duration at 100% saturation duty cycle using the pTx-CEST sequence (second row) compared to those using the conventional non-pTx-CEST sequence (first row) under the TR of 3s (first column), 4s (second column), and 5s (third column). For both pTx- and non-pTx-CEST sequences, each pulse element was 35ms, and the second number at the top of each image referred to the maximum allowable number of elements.

Figure 4. CEST source images obtained at +3.5 ppm (first column) and -3.5 ppm (second column) in conjunction with APTw images (third column) from the abdomen of a volunteer using the pTx-CEST sequence with the optimal setting manifesting elliptical polarization (second row) compared to circular polarization (first row).