Figure 1: 4D PC MRA visualization of a subject participating in this study using the CAAS software (Pie Medical Imaging). A) Based on the fully automated centerline detection, cross-sections through the different Internal Carotid Artery (ICA) segments, C1-C7,were selected as illustrated by the 7 slices, with matching numbers from 1 till 7. B) Visualized streamlines through the selected planes.

Figure 2: Boxplots showing the median, interquartile range, and minimum-maximum values of the variation in A) velocity Pulsatility Index (vPI); B) Arterial Distensibility (mmHg^-1) for patients with cerebral small vessel disease (CSVD) and controls per internal carotid artery (ICA) segment.

Figure 1. Data analysis workflow to quantify lenticulostriate arterial (LSA) blood flow velocities (BFVs). (A) The reference scan magnitude images after an N4 bias correction were threshold-filtered to generate vascular mask. (B) The velocity dataset was calculated from phase-difference images in three directions (Vx, Vy, Vz). (C) A mask was applied, and the velocity map was reconstructed with maximal intensity projections (MIPs). (D) Rectangular regions of interests (ROIs) with fixed sizes were placed on the MIP images to measure LSA BFVs.

Figure 2. A representative case of symptomatic patient with cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) showing reduced lenticulostriate arterial (LSA) blood flow velocities. Velocity map of (A) a CADASIL patient; (B) an age- and sex-matched asymptomatic NOTCH3 mutation carrier; and (C) an age- and sex-matched healthy control.

Figure 1: (A) Rapidly switching electric currents in the gradient coils cause mechanical deflection according to Lorentz’ law. The deflection causes the magnetic field gradient to oscillate with mechanical resonance frequencies. These can be identified (B) using either the Gradient Modulation Transfer Function (GMTF) or the Relative Microphone Amplitude Spectrum (RAMS). Of note, the RAMS during scanning can be acquired without special equipment (audio recording with mobile phone).

Figure 4: Comparison of TR=5ms (‘good’) and TR=4.69ms (‘bad’) for TE=2.35ms and $$$f_{MEG}=1065$$$Hz. Phase images are shown with profiles along frequency- and phase-encode direction (FE/PE) and corresponding linear fit (red line). For “bad” TE, an increased slope is seen, which is quadratic in the FE direction (arrows). As predicted for inter-TR effects, flow encoding in one axis influences other axes. Of note, X and Y gradient axes (Golay pairs) show similar behaviour due to similar mechanical properties.

Figure 2. Velocity differences for increasing acceleration factors in one representative subject. A. MIP of the velocity encoded in the z direction (cm/s) for each acceleration factor (a=[1,1.25,1.67,2.5,5]) and reconstruction method (5D flow at end-expiration and 4D flow fNAV). B. Voxel-wise velocity bias (in cm/s) in Bland-Altman plot between 4D flow fNAV and 5D flow at (a=1). C. Bland-Altman plots of the voxel-wise velocity differences between datasets with (a=1) and (a=[1.25,1.67,2.5,5]) for both 5D flow and 4D flow fNAV.

Figure 1. Study outline for reconstructing a free-running flow dataset. A. Reconstruction of motion-resolved 5D flow data uses the motion information to reconstruct images at different respiratory and cardiac phases. B1. Cardiac motion-resolved and respiratory motion-corrected fNAV 4D flow reconstructions take advantage of the respiratory motion and of data derived PCMRA images to iteratively estimate the rigid displacement of the heart due to respiration for every readout (B2) and use auto-focusing to correct this displacement to the end-expiratory position(fNAV)¹⁰.

Hemodynamic parameters obtained from 4D flow MRI with and without self-gating. In this figure we show visual representations of the hemodynamic parameters for one volunteer.

Bland Altman plots of velocity and wall shear stress at each section. The difference in Bland Altman plots is computed as: SG—NG, where SG is the data with self-gating and NG is the data without self-gating.

Figure 1: Summary illustration of the steps of the proposed stochastic 4D Flow signature technique. From left atrial 3D segmentation, the signature is computed from millions of pairwise 4D flow MRI’s 3D velocity-field vector disparities throughout the 3D left atrium volume and over the entire systolic time frame. This allows the signature to comprehensively and quantitatively characterize patient-specific complex LA flow dynamic alterations, including the interactions between various changing flow components.

Figure 3: (a) Individual 4D Flow signature profile of all 10 healthy controls. (b) The median and interquartile range of 4D Flow signatures for all controls (black, N=10) and all AF patients (red, N=30). (c) Box plot of the derived Hemodynamic Signature Index (HSI) values for controls and AF patients. P-value represents the results of the two-way t-test.

Detection of rTOF with preserved EF: ROC curves comparing the diagnostic performance of RV direct flow, efficient index and RV EF with respective AUC values. rTOF: repaired tetralogy of Fallot; EF: ejection fraction; RV: right ventricle; efficiency index: effective cardiac index/RV systolic KEi_EDV; KEi_EDV: kinetic energy normalized to end-diastolic volume.

Demographics, left ventricular (LV) and right ventricular (RV) flow analysis parameters in controls and repaired tetralogy of Fallot (rTOF)

Figure 2: Flow compartment visualization in a typical example before and after sildenafil administration. The fraction of right ventricular pathlines in the direct flow compartment (green pathlines) increases after sildenafil administration (white arrows). Direct Flow: Blood that enters the ventricle and leaves in the same heartbeat. Retained Inflow: Blood that enters the ventricle during diastole and stays for >=1 beat. Delayed Ejection Flow: Blood that starts in the ventricle and leaves during systole. Residual Volume: Blood that resides within the ventricle for >=2 beats.

Figure 1: Overall study design. On the first visit, each subject was randomly assigned to receive either intravenous metoprolol tartrate or oral sildenafil citrate. Each subject received the other drug on the second visit, so that all 9 subjects received both drugs. On each visit, the subject received 2 cardiac MRI scans: one baseline and one after drug administration.

Figure 2. The (a) peak diameter index, (b) regurgitation fraction, (c) systolic kinetic energy and (d) systolic viscous energy loss in 14 planes of normal, rTOF1, and rTOF2 groups. *p<0.05, **p<0.01, ***p<0.001, rTOF1 and rTOF2 patients compared to normal volunteers. ^†p<0.05, ^{† †} p<0.01, ^{† † †} p<0.001, rTOF1 patients compared to rTOF2 patients.

Table 1. Demographic characteristics of the study population.

Figure 3. Maps for abnormally elevated velocity (top) and abnormally directed WSS (bottom) in the Marfan cohort. Maximum incidence is delineated in the white circles.

Figure 2. a) Maps showing abnormally elevated velocity and WSS (a) are created by comparison with the control averaged and SD velocity and WSS maps. Regions of interest 1-6 are indicated in the WSS maps. b) Maps showing abnormally directed velocity and WSS are created by comparison with the control averaged velocity and WSS maps.

Fig 2.Examples of the LV endocardial contours at end-diastolic phase and flow components in a repaired Fontan (rFontan) patient (9-year-old boy). rFontan patient had reduced direct flow, increased residual flow than normal volunteer.

Fig 3.Examples of the LV endocardial contours at end-diastolic phase and flow components in a normal volunteer (11-year-old boy).

Figure 3: A. Kinetic energy ventricular vascular coupling (VVC), B. vorticity ventricular vascular coupling.

Table 1: Left ventricular flow analysis. where *p<0.05, KE is kinetic energy, mJ is millijoule, SV is stroke volume, EDV is end diastolic volume, L is liter, s is second, EL is energy loss

Figure 1: A) Workflow for segmentation and registration of 4D flow and CE-MRA data. From left to right: the LA was segmented from the 4D flow PC-MRA and the LA and LAA were segmented from the CE-MRA. The left atria are registered together, and the resulting transform is applied to the LA and LAA of the CE-MRA. Flow quantifications (stasis, peak velocities) are then quantified. B) Sample of the RR signal extracted from a continuously acquired real time phase contrast sequence and RR histogram

Figure 3: Box plots illustrating differences in mean stasis (top row) and peak velocity (bottom row) for the LA (left) and LAA (right) between low (n=21) and high (n=27) heart rate variability groups. *p<0.05

Figure 1: Pulsatility index and distensibility in PXE patients (blue) and healthy controls (red). Difference in A) pulsatility index and B) distensibility in healthy controls and PXE patients in different segments of the ICA and MCA. Lower panel: spaghetti plots of C) pulsatility index and D) distensibility to visualize consistency in behavior over the segments for the individual measurements.

Figure 2: Differences in pulsatility attenuation between PXE patients (blue) and healthy controls (red) between A) the C4 segment of the ICA and the MCA, B) the C4 and C6 segment of the ICA and C) the C6 segment of the ICA and the MCA. Although PXE patients showed slightly less pulsatility attenuation between C4 and C6 the net attenuation between

Figure 2. Correlation between MCA pressure increase ratio and BFVtotal increase ratio in three groups according to pre-MCA pressure levels. Dotted line represents the linear regression line of the patients with severe decrease of pre-MCA pressure (≦ 32mmHg). BFV: Blood flow volume, MCA: middle cerebral artery

FIgure 1. BFV of each artery (A, ipsilateral ICA; B, contralateral ICA; C, BA; D, ipsilateral STA; E, contralateral STA) and total BFV (F) before and after bypass surgery. *p < 0.05. BA: Basilar artery, BFV: Blood flow volume, ICA: Internal carotid artery, STA: Superficial temporal artery

Blood Flow Volume in Each Cardiac Phase.

The peak systolic blood flow volume was significantly decreased after EVAR (p=0.007), and the peak end-systolic reflection flow volume was significantly reduced (p=0.043) In the infrarenal abdominal aorta.

In the common iliac arteries, peak systolic blood flow was significantly increased (right: p=0.016, left: p=0.016). No significant decrease was observed in the peak end-systolic reflection blood flow volume.

Sections for Measurement.

The blood flow volume was measured in the sections at suprarenal and infrarenal abdominal aorta and both right and left common iliac arteries before and after stenting.

Figure 2. Velocity and wall shear stress (except Volume) decreased with age (p < 0.05).

Figure 3. Volume, velocity, WSS(max), WSS(mean) changes at different anatomical locations.

Figure 1. Streamline visualizations (color-coded to blood speed) in all 8 subjects. Helical flow (red arrows) was observed in all subjects.

Figure 3. Superior vena cava pathline visualizations in 4 volunteers at three cardiac timepoints each, showing (from left to right) no helical flow, matured helical flow, and disrupted helical flow, respectively. Type 1 (twining) is shown in the top row, and Type 2 (untwining) in the bottom row.

Figure ３. REPI 4d MRA of VENC=30cm/s (upper row). REPI can be visualized in a well-balanced manner from the ostial to distal portion of the left anterior descending artery.

Figure 5. REPI 4D MRA can reconstruct oblique images along the left anterior descending artery (arrow).

Fig.1. Nine planes were placed in different locations of the intracranial arteries (a). Velocity was defined as the maximum velocity of all voxels. The 20 time points represent different times of cardiac cycles (b).

Fig.2. Volume, velocity, WSSmax and WSSmean changes at different anatomical locations of young secondary hypertensive patients and healthy adults.

Figure 1.

50s woman with repaired TOF. VENC 50cm/s images can emphasise slow venous flow like collaterals without contrast medium (upper row). Arterial-pulmonary collaterals from the right subclavian artery were detectedas well as angiography (left lower). VENC 200cm/s images show severe stenosis at the proximal right PA (lower row).

Figure 2.

4D flow with VENC 200cm/s shows typical flow pattern of PA stenosis and aneurysm.

Figure 1 A 22-year-old female volunteer, reconstructed image (A), flow (B), and direction (C) of portal system were shown. Red lines representthe SV and yellow lines the SMV contributions to the total portal blood flow (D), respectively.

Table 1Comparison of flow volume measurements between fasting and meal

Figure 1: Example of AV measurements. 4D-flow: A) Color-coded streamline image with corresponding AV measurement plane (red line) (top left). B) Magnitude and C) Phase contrast images of AV measurement plane in 4D-flow dataset during systole and D1) & D2) corresponding MPR images. 2D-PC: E) Phase and F) Magnitude images of AV measurement plane

Figure 3: Boxplot (left) and Bland Altman plot (right) of measurement times between our clinical 4D-flow protocol and the 2D-PC series. All patients underwent phase contrast series for both aortic and pulmonic valves.

Figure 1. 3D segmentation at the level of the abdominal aorta (upper left) and corresponding 4D flow particle tracing from a 68-year-old man (lower left). Representative flow curves in the abdominal aorta, common hepatic artery and proper hepatic artery (right).

Table 2. Statistical analysis between pre- and post-TACE for abdominal aorta, common hepatic artery, and proper hepatic artery.

Figure 1: Schematic of computational method enhancing 4D Flow MRI for cardiovascular flow dynamics analysis. Top left: COA patient 2 anatomy pre-repair from CE-MRA. Bottom left: 4D flow MRI study of patient 2 before stent placement. Top center: Autonomously changing computational mesh grid over one cardiac cycle due to AMR. Bottom right: Flow information at ascending aorta extracted from in vivo study used for inlet boundary condition for numerical simulation. Top right: CFD simulation results showing flow information of patient 2 with aortic kinking.

Figure 5: Comparison of in-vivo 4D flow MRI and CFD results for post-interventions cases with healthy control. Good agreement between in-vivo and CFD results in velocity magnitudes at acceleration regions and vortical locations can be seen. Results show residual complex flow regions with larger vortices at ascending aortic section after repair when compared to the healthy case.

Figure 1: 4D Flow-derived metrics of flow inefficiencies correlate with simulated afterload in post-ASO DTGA. Maximum systolic energy loss (A) and maximum wall shear stress (B) are significantly correlated to pressure differential (ΔP) in the MCS circuit (p = 0.021, r = 0.57 and p <0.001, r = 0.85, respectively). Streamline visualization shows a range of flow patterns in these patients, where efficient flow (C) has relatively low energy loss, and inefficient flow (D) demonstrates relatively high energy loss.

Secondary Figure: In post-ASO DTGA patients, 4D flow CMR reveals inefficient pulmonary arterial flow patterns and quantification of flow inefficiencies by maximum systolic energy loss (mW/m3) reveals a significant correlation to pressure differential (ΔP) in mock circulatory system simulations (p = 0.021, r = 0.57). Representative segmented pulmonary arteries with relatively low and high energy loss shown.

Figure 3. Comparison of reverse flow metrics between healthy controls and BAV patients revealing increased reverse flow and reverse flow volume in BAV patients. Asterisk indicates p-value < 0.05, red crosses are outliers, and upper/lower bars denote the range within 1.5*interquartile range. (a) Mean reverse flow in the AAo, (b) mean reverse flow in the arch, (c) mean reverse flow in the DAo, (d) mean volume above 0.03mL/cycle, and (e) mean fraction of the AAo occupied by reverse flow volume.

Figure 1. Analysis Work-flow. Centerlines were automatically calculated for each segmented aorta (a) and planes were placed to select the regions of interest (ROIs) (b). Orthogonal planes were placed along the centerline for voxel-wise reverse flow calculations (c) to generate reverse flow maps and mean intensity projections for each patient (d). The threshold of 0.03mL/cycle was applied to each map to calculate reverse flow volume (e).

Figure 1 Segmentation image (A). Visualization module (B), placed measuring planes at the proximal, middle and distal of the PV, perpendicular to the vascular section, flow and WSS were measured. Functional module (C), placed reference plane at the origin of the PV, measuring planes were placed at the proximal, middle and distal of the PV to measure the pressure.

Table 1 Measured parameters of portal system with different CS

Figure 4 - (A) : Box-and-whisker plots of Flow_avg on bilateral sides in cortical infarct pattern (n=12). There is significant difference between Flow_avg of the infarction side and Flow_avg of the contralateral side.

Figure 1: 4D-flow data analysis using dedicated software. Flow pattern visualization was performed by streamlines (A) and hemodynamic measurements within contours (B, C) were used for quantification of blood flow.

Figure 1: (A) VEG waveforms used in simulation and corresponding spectra. (B) The overlap of motion spectra and VEG spectra defines the encoding power of the VEG. Depending on the eddy frequency$$$\;f_{eddy}\;$$$, VEGs can exhibit insufficient spectral coverage.

Figure 3: (A) Simulation results for VEGs of different frequencies. While mean velocity is estimated without bias, RST values are underestimated depending on VEG frequency. Increasing the VEG frequency reduces underestimation of RST values. (B) Profiles along FOVx and FOVz for estimated RST vs. Ground Truth.

Figure 2: A comparison of masks and flow curves in the AAo and MPA for one volunteer before and after phantom/second-order correction. MSAC shows superior performance by avoiding wrap-around regions compared to the static tissue mask (a). This leads to a correction of the flow curves (b) similar to the phantom correction and consequently to improved flow volumes and Q_P/Q_S-ratio (c).

Figure 5: Boxplot showing the RMSE performance over all acquisitions. While static fit corrections are beneficial in most cases compared to no correction, MSAC shows best median and interquartile performance over all cases. The MSAC outliers correspond to two images, both with severe wrap-around leading to more wrap-around pixels than no wrap-around.

Figure 2. Velocity was measured by unwrapping the velocity of the low-venc and using the high-venc phase image as an unwrapping threshold. A) Phase image from a high-venc acquisition with a low VNR. B) Phase image from low-venc acquisition with a high VNR and velocity aliasing. C) Magnitude image averaged from both echoes. D) Combined phase image with the VNR of the lower venc image without the velocity aliasing artifacts.

Figure 1. Sequence diagram of a retrospective ECG-gated 2D cine dual-venc dual-echo PC MRI. High and low-venc data were acquired within a single TR to minimize the acquisition time associated with the additional venc.

Representative streamline reconstructions from different measurements on Phantom-B with varying undersampling factors using the L1- and L2-ESPIRiT algorithm. The structure and density of streamlines were obviously preserved by the L1-ESPIRiT reconstruction in comparison to L2-ESPIRiT.

2D embedding of pairwise dissimilarities using MDS for velocity, vorticity, and helicity density fields, respectively, for Phantom B. The pairwise Euclidean distances between points resemble pairwise field dissimilarity. The axes are dimensionless and only depict the relative volumetric dissimilarity normalized to the maximum dissimilarity value.

Figure 2: EPI and CS 4D flow MRI streamline visualizations in four different patients with: pulmonary valve regurgitation (top left), aortic valve regurgitation (top right), tricuspid valve regurgitation (bottom left) and mitral valve regurgitation (bottom right). Semi-automated retrospective valve tracking was performed on bSSFP cine images, on two orthogonal views for each heart valve. Rvol = regurgitant volume.

Figure 1: EPI and CS 4D flow MRI streamline visualizations of blood flow through the aortic valve (yellow contour), mitral valve (orange), pulmonary valve (blue) and tricuspid valve (green) in a 28-year old healthy volunteer, resulting from semi-automated retrospective valve tracking. Valve tracking was performed on bSSFP cine images, on two orthogonal views for each heart valve. 4-chamber bSSFP view is visible in the background.

In A, the 4-chamber is shown (LV: left ventricle, LA: left atrium, RV: right ventricle, RA: right atrium). In B-D, gradient orientations for 4DEPI and 4DGRE are shown, similar as for the LV phantom. 4DEPI1_LA: long-axis oriented EPI with readout gradient parallel to the main flow, 4DEPI2_LA: long-axis oriented EPI with blip phase encoding gradient parallel to the main flow, 4DEPI3_SA: short-axis oriented EPI with both readout and blip phase encoding gradients perpendicular to the main flow. A cylinder-shaped control volume is positioned below the mitral valve.

In A, 1 indicates the LV phantom, 2 and 3 indicate bioprosthetic mitral and aortic valve. In B-D, gradient orientations for 4DEPI and 4DGRE are shown. 4DEPI1_LA: long-axis oriented EPI with readout gradient parallel to the main flow, 4DEPI2_LA: long-axis oriented EPI with blip phase encoding gradient parallel to the main flow, 4DEPI3_SA: short-axis oriented EPI with both readout and blip phase encoding gradients perpendicular to the main flow. A cylinder-shaped control volume is positioned below the mitral valve.

Fig 1. Siemen’s FIRE framework allows for augmentation to Siemen’s image reconstruction environment (ICE), creating an interface where data can be requested and sent back into the ICE pipeline via FIRE emitter and injector. Reconstructing 4D-Flow images are sent to a containerized Python environment. A 3D Phase Contrast MRA is calculated prior to input to turnkey execution of our networks (CNN, outputs above). An aortic MIP cine is calculated from the velocity data and aortic segmentation and sent back to the ICE pipeline to be delivered to the console with reconstructed 4D-Flow data.

Fig 4. 4D-Flow with integrated aortic velocity MIP cine visualization using FIRE in a healthy control. The processing pipeline included deep learning pre-processing and segmentation with calculation of an aortic velocity MIP cine. The MIP cine is displayed on the console alongside the standard reconstruction of phase and magnitude images. Deep learning processing and segmentation performed successfully despite artifacts from a metallic spinal implant.

Figure 3. Quantification results of flow and momentum along the jet axis (voxel size 1.5 mm³ results) A: Flow rate waveform vs. x [mm] along the jet (see Fig.1 for details). B: Flow momentum waveform vs. x [mm] along the jet. C: Total jet flow volume vs. x [mm] along the jet . The right vertical plot axis indicates the error between RVol by 2D PC MRI. D: Momentum-Time-Integral vs. x [mm] along the jet. Regression statistics were acquired for x = 5-50 mm.

Figure 5. Relative voxel size dependency of flow quantification. A: Difference in flow volume compared to voxel size 1.5 mm results vs. number of voxels across the jet cross-section. B: MTI vs. number of voxels across the jet cross-section. The jet diameter was derived by computing the diameter of a circle with the area equivalent to the peak jet area.

Figure 2. (Animated GIF) Example time-resolved segmentations in subjects from Figure 1, generated automatically through non-rigid registration of phase-contrast MR angiographic images at each cardiac time point to a reference volume. Inset: time-resolved contour in the ascending aorta and resulting flow waveform over 60 cardiac frames.

Figure 2. (Animated GIF) Example time-resolved segmentations in subjects from Figure 1, generated automatically through non-rigid registration of phase-contrast MR angiographic images at each cardiac time point to a reference volume. Inset: time-resolved contour in the ascending aorta and resulting flow waveform over 60 cardiac frames.

Scatter plots demonstrating correlation between pre- and post-contrast Total Forward Volume (TFV) and Peak Velocity (PV) measured by the conventional and CS 4D flow techniques.

Bland-Altman plots representing the agreement in pre- and post-contrast Total Forward Volume (TFV) and Peak Velocity (PV) between the conventional and CS-based 4D flow techniques. Blue solid lines show the mean of differences, while the dotted lines indicate the upper and lower limits of agreement (±1.96 SD).

Fig. 2 (a) Phase data for four-point Cartesian and ICOSA6 100%. Smoothing effects were observed in ICOSA6 data, particularly where helical flow with a large velocity range is present. Four-point Cartesian included aliased pixels prior to the model inlet, which was not present in ICOSA6 data. (b) End-diastolic pathlines based on four-point Cartesian and ICOSA6 data. With increased under-sampling, the pathline travelling range decreased, the detection of helical flow in the proximal FL becomes more challenging, and jet flow velocities (entry tear) become smaller.

Fig. 3 (a) TBAD cross-sections. (b, c) Calculated net flow and (d, e) peak velocity for four-point Cartesian and ICSOA6. Conservation of mass dictates that net flow through ‘inlet’ versus ‘BCT, ‘entry tear’ versus ‘exit tear’, all TL sections, as well as all FL sections should be equal. Flow waveforms shown for the entry tear (f), true lumen (g), and false lumen (h). Overall, flow waveforms between the four-point Cartesian (blue) and ICOSA6 100% (green) as well as 50% (red) correspond well. With increased under-sampling of ICOSA6 reconstructions, peak flow rates decrease by up to 49.1%.

Figure 1. Example ECG, SG, and PT signal traces with corresponding cardiac triggers from a healthy volunteer (Volunteer 1). A 30 second window is shown for clarity.

Figure 2. Bland-Altman plots comparing net volumetric flow, Q, quantified from the ECG-, SG-, and PT-binned 4D flow reconstructions with respect to the 2D-PC reference. Biases and limits-of-agreement (LOAs) were computed from the aggregate of Aao and MPA measurements. The y-axis is in terms of milliliters.

Images obtained by a micro CT unit (SKYSCAN1176, Bruker)

a. A front view. b. A side view. c. A tilted view of one layer.

Comparison of (a) signal intensities of phantom imaging (b = 50 s/mm², MPG of the phase encoding direction) with (b) those obtained by computer simulation. The intensities are normalized to those with the infusion rate = 0.

Figure 1: (a) The overall architecture of proposed architecture. Back to back residual block and attention block is adopted as the backbone of a U-Net. Each residual attention (RA) block is followed by maxpooling in encoder network and upsampling and concatenation in decoder network. (b) shows the architecture of a RA block. Each residual block consists of 3 convolutional layer followed by batch normalization and ReLU. Attention block consists of channel attention and spatial attention.

Figure 2: (a) & (b) shows magnitude image and Phase Image at FH, AP and RL direction from reference complex image, undersampled complex image, image reconstructed by U-net, TV regularization and image reconstructed by proposed method for two subjects. The image is in the peak systole phase of the cardiac cycle and is exactly at the location of the aortic valve for both subjects.

Figure 2: (a) Pre- COA intervention in-vitro silicone model used for flow 4D flow MRI and PIV experiments. (b) Hemodynamic velocity maps obtained by 4D flow MRI with an average inflow rate of 2.6 LPM. Enhanced visualization of complex flow features in the mycotic aneurysm are observed.

Figure 4: (a) Tomographic (3 camera) PIV setup. A laser is pulsed through a plane cutting a cross section of the model and synced with the acquisition of the high-speed camera. (b) Visualization of particle traces seeded in the aortic flow, illuminated by the laser. The high-speed cameras track the displacement of the particles over the cardiac cycle which are used for velocity quantification. (c) Preliminary velocity map at a cut-plane at the center of the aortic model at peak systole. The high velocity jet extends from the aortic arch to the aneurysmal portion of the descending aorta.

Experiments results of transvalvular pressure drop (ΔP) measured invasively via peak-to-peak and non-invasively via simplified advective work-energy relative pressure (SAW) and simplified Bernoulli (SB). CO1, CO2 and CO3 represent respectively the pulsatile with a maximal flow rate of 150ml/s, 200ml/s and 250ml/s.

Phantom set-up representation, with the illustration where the valves were implemented.

TKE mapping in 4D Flow & ICOSA6 with CS reconstruction.

Velocity mapping in 4D Flow & ICOSA6 with CS reconstruction.

Figure 3: The mean signal values from the center region are shown in dependency of flip angle (a) and slice thickness (b) for different pump levels for numerical simulation (dashed) and experiment (solid). The error bar shows the standard deviation. The left figure is for a slice thickness of 6 mm and the right is for a flip angle of 20°.

Figure 2: The numerical (dashed) and experimental (solid) velocity (a) and signal (b) profiles for different pump-levels are plotted for the intersecting line inside the pipe. The region in the center is used in further comparisons (gray).

Figure 3) A microrobot (A, red circle) in the 60 degree bifurcation PVA phantom stuck in an eddie visualized using ink (B, orange arrows) and 4D flow (C) (0.4x0.4mm² in plane resolution). The eddies were simulated with streamlines (D).

Figure 4) Measurement of the flow velocity according to the resolution of acquisition and the cross section without and with resistance in the branch (B1, left).

Figure 1. Flow metric estimation showing quantitative representation between aortic flow sequence (dotted lines) and Whole heart sequence (hard lines) across different subjects.

Figure 2. Flow metric estimation of velocity and volume between aortic flow sequence (dotted lines) and Whole heart sequence (hard lines) across different subjects.