Figure 2. Scatter plots of four imaging parameters for different pathohistologic outcomes: NASH diagnosis, steatosis, inflammation, cellular injury, and fibrosis. *:P-value<0.05 is considered statistically significant.

Figure 4. Nominal logistic regression with effect likelihood and odds ratio tests for three imaging parameters. *:P-value<0.05 is considered statistically significant.

Fig.1. Tri-exponential IVIM fit of the liver using IVIM₃-NET (NN) versus nonlinear least squares (LSQ) fit in three non-alcoholic fatty liver disease patients with progressive fibrosis grades. Parameters fit for both fit methods were diffusion (D), slow pseudo-diffusion (D*₁), fast pseudo-diffusion (D*₂), slow perfusion fraction (f₁) and fast perfusion fraction (f₂). The IVIM₃-NET shows less noisy parameter map images when compared to LSQ-fit. Particularly for f₂, a decrease in signal is visible as fibrosis levels increase.

Fig.2. Spearman correlations between histology and IVIM parameters fitted using IVIM₃-NET (NN) and nonlinear least squares (LSQ) fit. Crossed out correlation coefficients represent non-significant correlations, others all had a p-value < 0.05. While performance was similar, the IVIM3-NET showed slightly more and stronger correlations between IVIM parameters and histology. Overall, fibrosis showed the most and strongest correlations with IVIM parameters, followed by inflammation and ballooning resp. Steatosis levels did not correlate with any IVIM parameters.

Figure 1. Example MRE wave images, elastograms, MRI anatomic and upper endoscopic images in two patients; one from high-risk group (A) and the other from low-risk group (B).

Figure 2. Receiver operating characteristic (ROC) curves of the training and validating cohorts showing the ability of ordinal logistic models to grade patient’s variceal severity. The predictors included in each model are demonstrated.

Figure 1. Example MPF maps, a): from a patient without liver fibrosis (F0), and b): from a patient with fibrosis stage F2. All results were confirmed by liver biopsy.

Figure 2. The box plot of the measured MPF using MPF-SL at stage F0 and stage F1/F2.

Figure 3: Qualitative comparison of zero-filling, POCS and DRPF-Net for a combined DW liver image (b = 800 s/mm²). First row: the zero-filled reconstruction is blurred along AP, while POCS exhibits artificial, ringing-like structures most apparent in the left liver lobe. In contrast, the DRPF-Net restores image sharpness without introducing visible artifacts. Second row: difference (5x) to the ground-truth. Third row: phase maps of a selected repetition with phase estimation errors coinciding with the locations of reconstruction errors in the magnitude images.

Figure 1: POCS reconstructions of retrospectively PF-sampled liver images (b = 800 s/mm²) in two representative cases: a) one with relatively smooth phase and b) one with high-frequency phase fluctuations. In a), the phase can be approximately restored using the LR phase estimate shown in the bottom right. However, ringing (white arrows) as well as a grid pattern (orange arrow) are still introduced into the magnitude image. The artifacts in b) are much more severe leading to signal-dropout in the liver (red arrow) whose location coincides with wrong estimations of the phase.

Fig.1 The framework of 4D-DW-Propeller-EPI technique with golden angle. It mainly includes two parts: blade data acquisition and 4D-DWI data reconstruction. Abdomen was scanned with 180 blades for each type of two diffusion contrast (b = 0 s/mm² and b = 400 s/mm²). The recorded respiratory signal was used for sorting the blade data and placing them into corresponding phase bins. The blades falling into the same bin were combined using Propeller-EPI reconstruction.

Fig.4 Left panel shows the In-vivo 4D-DW-Propeller-EPI images reconstructed from data acquired with (a) TR=1250ms, and (b) TR=2000ms. Right panel shows the number of blades used for reconstructing six respiratory phases for the usage of 100%, 75%, 50%, and 25% of all acquired blade data (180 blades). Shortening sampling duration can significantly reduce the number of blades available in each phase bin for subsequent Propeller-EPI reconstruction, resulting in degraded image quality. The reduction of blade number in different phases is not the same, leading to varying image quality.

Figure 4. (A) Representative RT-DWI and ADC map (a), DT-DWI and ADC map (b), MoSe-DT-DWI and ADC map (c) on a 1.5T scanner. In the RT-DWI and DT-DWI, signal loss from cardiac related artifacts appeared and the ADC left was artificially higher than ADC right (arrows). MoSe-DT-DWI significantly improved the image qualities of the left liver lobe compared to other techniques. (B) Representative RT-DWI and ADC map (d), DT-DWI and ADC map (e), MoSe-DT-DWI and ADC map (f) on a 3T scanner, similar findings to 1.5T were obtained.

Figure 5. The SIRs and ADC ratios were shown. The SIRs and ADC ratios at the MoSe-DT-DWI is the closest to one, suggests that MoSe-DT-DWI is the least impact of signal loss from cardiac motion.

Fig.3 The representative liver diffusion images (b=0 and b=500 s/mm²) and corresponding ADC maps generated from three routine liver DW-EPI methods and free-breathing DW-Propeller-EPI. Compared to three routine liver DWI methods, the DW-Propeller-EPI shows better geometric accuracy (by comparing the liver contour obtained from FGRE image), reduced geometric distortion (yellow arrows), and less image blurring. The DW-Propeller-EPI is based on multi-shot acquisition, and therefore lengthens the scantime (i.e., 5:40).

Fig.5 The evaluation of repeatability of ADC measurements of all sequences using the methods of Bland and Altman. The free-breathing liver DW-Propeller-EPI shows best repeatability in ADC measurement, with better ICC (0.912) and lower CV (5%) than other three routine liver DW-EPI methods, suggesting that DW-Propeller-EPI may be more suitable for cross-sectional or longitudinal liver DWI application.

Figure 3. Peak arrival times of total ¹³C signal for each subject (t = 0 coincides with the start of the acquisition). The tumor for Cancer #2 fell within a saturation band and no signal was acquired (marked with a diamond).

Figure 1. Total HP ¹³C signal plotted over time for selected tissue voxels in three healthy human subjects with ROIs labeled in the adjacent anatomical image. Spatial resolution varied between acquisitions.

Fig. 4. Example of the utility of high frame rate reconstruction. Adequate scan timing of arterial phase (AP) was not observed on routine frame reconstruction (12 s/phase). Portal vein (arrowhead) was not enough enhanced in the 3rd phase and hepatic vein (dotted arrow) was already enhanced in the 4th phase. Optimal AP was obtained with high frame rate reconstruction (3 s/phase). Portal vein (arrowhead) was enhanced but the hepatic vein (dotted arrow) was not yet enhanced in the 13th and 14th phase.

Fig. 5. Stacked bar graph representing results of the visual assessment. No respiratory motion/pulsation artifact was observed in DISCO-Star dataset (P < 0.0001). DISCO-Star showed lower score of streak artifact and overall image quality (P < 0.0001). However, there was non-inferiority in the proportion of diagnosable image in DISCO-Star dataset in comparison with DISCO dataset.

Figure 1: 13-year-old boy with autoimmune sclerosing cholangitis. A) Iron-corrected T1 (cT1) map of the liver. B) MR elastogram of the liver (passive driver placed over right upper quadrant). C) MR elastogram of the spleen (passive driver placed over left flank).

Figure 2: One-way ANOVA (mixed model) results to determine statistical differences in measurements of interest between baseline, year 1, and year 2 exams for AIH and PSC / ASC cohorts. P-values reported. AIH: autoimmune hepatitis; PSC: primary sclerosing cholangitis; ASC: autoimmune sclerosing cholangitis; ROI: region-of-interest; cT1: iron corrected (T2*) – T1 mapping; MRE: magnetic resonance elastography.

No correlation between liver fat fraction and hypertrophy rate after PVE

Loose inverse correlation between age and weekly kinetic growth rate R = -0,24 (p=0,04)

Example of classic predominantly hypointense HBP of a I-HCA

Examples of atypical homogeneously hyperintense HBP of I-HCA and homogeneously isointense HBP of I-HCA

Time intensity curves obtained from the two volunteers.

(a) TIC from the first volunteer.

(b) TIC from the second volunteer.

Blue, Yellow, Green, Red lines represent TICs from input-21, input-105, recon-CNN and recon-CS, respectively.

Reconstructed images from the first test volunteer.

(a) A subset of the input images with 21 spoke per frame (input-21.)

(b) A subset of the input images with 105 spoke per frame (input-105.)

(c) Output images from the CNN (recon-CNN.)

(d) Reconstructed images with the CS algorithm (recon-CS.)

Rows of the image matrix represent different temporal frame.

Dashed circles indicate the regions that the SNRs are calculated.

(a) Schematic of two-trait predictor of venous invasion (TTPVI), arterial peritumoral enhancement, non-smooth tumor margin, and capsule appearance. (b) Frequencies of presence and absence of the four features at CT and MRI. The number of cases showing TTPVI, arterial peritumoral enhancement, non-smooth tumor margin and capsule appearance at MRI but not at CT was 15, 12, 7, 18, respectively; Number of cases presenting the four features at CT but not at MRI were 5, 3, 6, and 0, respectively. (c) Sensitivity and specificity of the four features for MVI diagnosis at CT or MRI.

Odds ratios of each variable contributing to the R and RR models at CT or MRI. R model = Radiographic model; RR model = radiographic-radiomics model; RS = radiomics signature; TTPVI = two-trait predictor of venous invasion.

Table 1

Table 2

Figure 1. Nomogram, including ALB, GGT, MVI (microvascular invasion), and BCLC stage, to predict early recurrence (ER) in patients with hepatocellular carcinoma after curative resection. The nomogram can help obtain the probability of ER (≤2 years) by adding up the points identified on the points scale for each variable.

Figure 4. Decision curve analysis (DCA) for predicting early recurrence. Pink line: All patients dead. Blue line: No patient dead. Red dashed line: Model of the nomogram. Yellow line: Model of the BCLC stage. Green line: Model of the AJCC-TNM stage (eighth edition).

Combined administration of anti-PD-L1 antibody and sorafenib shows enhanced anti-tumor effects. A: Flow chart of in vivo experiments. B: Response of the subcutaneous tumors to the indicated treatments. C: T2WI of four groups of subcutaneous tumors before and at different time points after treatment. D: Pseudocolour images (T1 mapping) of four groups of subcutaneous tumors before and at different time points after treatment. E: T1 values of four groups of subcutaneous tumors before and at different time points after treatment (P>0.05). *p < 0.05, **p < 0.01, ***p<0.001.

Water ADC maps of representative treated groups obtained before therapy and at different times post-therapy. A, B: After 7 days of treatment, the ADC value of the combined treatment group increased significantly compared with the control group.C: Hematoxylin and eosin (H&E)-stained histologic sections of representative treated groups obtained at 21d post-therapy. Data are presented as means ± SEM. *p < 0.05, **p < 0.01, ***p<0.001.

Figure 3. ROC curves of DCE, T1/T2mapping, and the combine.

Figure 1. A 61-year-old male with HCC. Ktrans map (a), Kep map (b), Ve map (c), Vp map (d), T1 mapping (e), T2 mapping (f), T2WI (g).

Figure 2. A 65-year-old female patient with HM. Ktrans map (a), Kep map (b), Ve map (c), Vp map (d), T1 mapping (e), T2 mapping (f), T2WI (g).

Figure 5

Figure 4

Decision tree illustrating major feature combinations and HCC proportions(%HCC)/counts according to LI-RADS v2018 in the training (A) and testing (B) sets. Data were computed based on sum interpretations of the three readers while adjusted with a generalized estimating equation model. rLI-RADS diagnostic table based on rLR 3-5 category definitions derived from the decision tree (C).

LI-RADS, Liver Imaging Reporting and Data System; rLI-RADS, revised Liver Imaging Reporting and Data System.

Figure 2. In-phase 2D-GRE (top left), opposed-phase 2D-GRE (top right), in-phase 3D-SPGR (bottom left), opposed-phase 3D-SPGR (bottom right). The right hepatic lobe lesion (arrows) demonstrates signal loss on opposed-phase images with both sequences, as rated by both readers. Respiratory motion was scored mild and severe by both readers for the 2D-GRE and 3D-SPGR images, respectively. Of note, each pair of IOP images was assessed at different time points within the full slate of their respective sequences.

Figure 5. Statistical analyses of assessment methods.

Figure 4. ROC curve of the values for predicting the recurrence of early stage of HCC treated by TACE. The area under curve of tumor size, MK, MTRasym and the prediction model of tumor size+MK+MTRasym were 0.673, 0.861, 0.728 and 0.967, respectively.

Figure 3. ROC curve of the values for predicting the postoperative hepatic insufficiency of HCC treated by TACE. The area under curve of MTRasym was 0.783 and the MKvalue was 0.811.

Table 1. Discriminative performance of different predictive models in the training and validation cohorts

Figure 1. ROC curves for the radiomics model, clinical-radiological model, and combined model in the training cohort (a) and validation cohort (b).

Figure 1. In rTSE imaging, hyperintense? signals originating from outside the imaging FOV could propagate into the imaging FOV, causing streaking artifacts (a). In the proposed method, a mask was first generated to cover the imaging object within the imaging FOV (b). A coil-mixing matrix was then calculated (c), which is used to process the k-space data before reconstruction. The streaking artifacts in both the composite images and the T2 maps could be effectively removed by the coil-mixing de-streak method (d,e).

Figure 4. The accelerated rGraSE acquisition offers comparable image quality and T2 maps as the rTSE but takes half of the time to acquire.

Table2 Correlation between quantitative parameters of IVIM and 4D flow

Figure 2 a 65-year-old man with liver cirrhosis. 2a-2d Quantitative parameter (sADC, F, D, D*) mapping of IVM.

Figure 1. Visualization of the Blood Flow (left) and the Plane/Contour Used for the Quantification of the Blood Flow (right)

Figure 3. Using Hemodynamic Parameters to Help Stratify the Risk of Variceal Bleeding: Normalized Flow_cyc (left) and V_avg (right)

Figure 1. Graphical abstract of the experimental protocol. Portal occlusion induced by portal vein thrombosis were modeled by successive inflations of a balloon catheter in the portal vein of four pigs. Portal peak flow (PF), peak velocity magnitude (PVM) and liver stiffness µ were then measured using 4D-Flow MRI and magnetic resonance elastography (MRE) for successive inflation states.

Figure 2. Results showed a dependence of liver’s shear modulus with portal PVM (A) and PF (B). The reduction of the portal venous blood flow resulted in a decrease of liver stiffness. Consequently, portal occlusion implied by portal vein thrombosis may attenuate the increase in stiffness due to moderate fibrosis and lead to false-negative diagnosis with elastography. Spearman’s r_s rank correlation coefficients close to 1 indicate individual monotonic relationships (p-values are indicated).

Figure 1. Flowchart of the study population. (a) the workflow of our study population before and after propensity score matching (PSM); (b) the frequency of MVI stratified by tumor size.

Figure 2. The Kaplan-Meier curves in terms of MVI statuses and grades. (a) and (b) were stratified by MVI statuses and grades in the Pre-PSM subgroup, respectively; (c) and (d) were stratified by MVI statuses and grades in the Post-PSM subgroup , respectively.

Figure 2: Box plots of the bi- and triexponential IVIM parameters. Each point indicates the data of one volunteer. The central mark of each box plot indicates the respective median, the bottom and top lines indicate the 25th and 75th percentile. The whiskers reach out to ±2.7 $$$\sigma$$$, outliers are shown as “+”. Dashed horizontal lines show the threshold where the scale of the y-axis is changed in order to display all points in a reasonable way.

Figure 1: Signal curves. For better readability, the displayed signal values represent the values averaged over diffusion directions and slices . (a) Representative normalized signals of one volunteer at TE = 45 ms. Both the biexponential and triexponential fit curves are shown. Data points with lower brightness were used for the triexponential fit only. (b) Representative absolute signal of one volunteer at four echo times. Lines represent the triexponential IVIM curves.

Figure 2. Example images of MONO, MODI and MODI-DL in another volunteer. The yellow arrows indicate the left lobe of the liver where MONO leads to severe shading and signal dropouts in DWI and bias (overestimation) in ADC maps. MODI is able to mitigate the signal dropouts with consistent ADC throughout the liver. However, MODI DWI had lower SNR compared to MONO. By combining MODI with the DL reconstruction, MODI-DL DWI shows higher SNR and is able to maintain excellent motion-robustness throughout the liver.

Table 2. ROI measurements of mean ADC, ADC standard deviation (STD) and ADC SNR calculated as mean ADC / ADC STD

Fig. 1 FROC-derived parametric maps of one patient with low stage (F0) of liver fibrosis.

Fig. 5 Receiver operating characteristic curves for prediction of liver fibrosis (A) stage 1 or greater, (B) stage 2 or greater, (C) stage 3 or greater and (D) stage 4 using the combined diffusion index (D, β, μ and ADC), APRI, and FIB-4. Numbers are areas under the curves with 95% confidence intervals in parentheses.

Figure 1. Representative images of conv-DWI and CS-DWI for group A and group B, which received comparable ratings.

Figure 2. Example of a patient with a missed FLL on CS-DWI.

In conv-DWI (A), a FLL was detected in segment V (arrow), that was not called on CS-DWI (B). In correlation with all sequences from the standard MRI protocol, this FLL corresponded to a thrombosed liver hemangioma.

Figure 2. Simultaneous multi-slice diffusion‑weighted images (SMS-DWI) and the corresponding apparent diffusion coefficient (ADC) maps from two breathing schemes and two vendors in a volunteer. The upper row illustrates SMS-DWI (a-d) with b value of 50 s/mm². The lower row shows ADC maps (e-h), which were automatically generated on the MR system's console. Three circle region of interest (ROIs) in the right liver lobe were firstly draw on SMS-DWI, and pasted to corresponding ADC maps.

Figure 4. Box and whisker plots illustrates the apparent diffusion coefficient (ADC) measured in the right liver lobe from two breathing schemes and two vendors. In vendor 1 (Siemens system), the liver ADC of free-breathing SMS-DWI was significantly higher than that of breath-hold SMS-DWI (P=0.03). In vendor 2 (GE system), no significant difference was found in ADC values from different breathing schemes (P=0.42). The liver ADC values from two vendors did not show significant difference in both breathing schemes (P=0.05 and P=0.50). BH, breath-hold; FB, free-breathing.

Fig.3. Average and standard deviation of the diffusion parameters fitted by different methods and with set of b values on real data acquired on a patient.

Fig.2. MAPE on the parameter estimation computed over simulated liver data for the different fitting method and set of b values (‘Full’ and ‘Selected’ by the Monte Carlo method), and diffusion models. Error bars correspond to standard error of the MAPE over the simulated liver.

Figure 2. Images in a 36-year-old man with HCC (arrow). A, Delayed phase image showing a heterogeneous low-signal intensity lesion in segment VI of the liver. B, Diffusion map showing lower signal intensity compared with that of the liver parenchyma. The mean diffusion value of the lesion was 1.35 x 10^-3 mm²/sec. C, Kurtosis map showing higher signal intensity compared with that of the liver parenchyma. The MK value of the lesion was 0.83. D, ADC map showing a lesion with a mean diffusion value of 1.18 x 10^-3 mm²/sec.

Table 1. Diffusion parameters in HCC between the Ki-67>25% and Ki-67≤25% groups

Figure 4. PGSE data with pF = 0.7: single average DWI (a) and ADC maps (b). Without phase correction both homodyne and POCS result in DWI with “worm-like” artifacts in the left liver lobe (red arrows) and with homodyne also around the inferior vena cava (IVC) (green arrows) (a). Phase correction removes the artifact and restore image quality (a). Corresponding ADC maps show an overestimation of ADC around the IVC when homodyne is used without phase correction (b). In the left liver lobe both homodyne and POCS without phase correction yield noisier ADC values than with phase correction.

Figure 1. The phase correction procedure before the partial Fourier reconstruction.

Figure 1. The processing flowchart. Both μ_MRE and ADC were processed with corresponding in-house prototype software developed by MATLAB. Same ROIs were calculated to find the relationship of μ_MRE and ADC for all patients.

Figure 2. Linear regression of μ_MRE and ADC calculated from different combinations. a) μ_MRE vs. ADC calculated from b = 200 & 500 s/mm²; b) μ_MRE vs. ADC calculated from b = 200 & 800 s/mm²; c) μ_MRE vs. ADC calculated from b = 200 & 1200 s/mm². (d) Mean ADC values for all 73 patients at different stage of liver fibrosis stages. Significant differences were observed at different stages.

Figure 1. Representative b=800 s/mm² images and ADC maps using SS-EPI and iShim methods.

Figure 2. Bland-Altman plot for the ADCs between SS-EPI and iShim sequences. The results indicating the high accordance of ADCs between two sequences.

Figure 5: The ADC maps for the motion compensated waveforms do not show as large of an ADC overestimation in the left liver lobe when compared with pgse. The subtraction maps therefore do not show as large of a difference between the gnl corrected and non-corrected ADC maps.

Figure 3: Phantom results for all waveforms. The non-corrected ADC for an ROI drawn in the centre of the phantom shows the characteristic gradient nonlinearity induced parabolic shape. After correction, this bias is reduced for all waveforms, even if the variation in the ADC of the motion compensated waveforms is larger than pgse.

Representative volunteer images. DWI of b = 800 sec/mm2 images and ADC maps between reference, MB-only and MB+VERSE.

Comparison of VERSE pulse and SINC pulse. VERSE pulse algorithm is a way to simultaneously reshape an RF pulse and its slice-select gradient waveform while maintaining the same excitation profile for on-resonance spins. In short, if some portion of an RF waveform is stretched in time, a corresponding reduction in RF amplitude must be made to maintain the same flip angle.

Table 1

Figure 1. The noise and clarity of the liver contour obtained by AIR Coil (Score=5) are significantly higher than those of Traditional Coil (Score=3).

Table 1 : Repeatability study results. In bold, the optimal repeatability coefficient.

Figure 1: Examples of resulting ADC maps, superimposed on magnitude images at b-value = 0 s/mm². A) using volume coil, B) using 4-channel parallel coil, C) without respiratory gating, D) with respiratory gating. Volume coil images have larger sensitivity area while surface coil images has higher signal magnitudes. Respiratory gating visually reduces signal artifacts.

Fig 1 AUC of T1_pre, T1_HBP (ms) and ΔT1% with Child-Pugh B+C

Table 2 The correlation among T1_pre, T1_HBP (ms) and ΔT1% with ICGR 15% and TE

Fig. 1. Typical histological sections of the control and MCD diet groups by Hematoxylin-eosin (HE) and Sirius red staining.

Fig. 2. Line graphs of T1 relaxation time (a) and reduction rate of T1 relaxation time of the liver (b) in different stage of fibrosis. (c) Adjusted P Value of Reductin rate of T1 relaxation time (%) F0+1 vs F2+3.

ROC curves for distinguishing between patients without CLD, and CLD and cirrhosis on the basis of T1 reduction rates at 1.5+3T are shown (A-D). AUC values for a given cutoff for each comparison as well as the sensitivity, specificity and p-values are given. Cutoffs were determined using the Youden Index, which is also shown.

Table 3 Diagnostic performance for T1pre, T1post, rrT1, HeF and K_Hep values in assessing ICG R₁₅＞20% groups by ROC analyses.

Table 1 Liver T1pre, T1post, rrT1, HeF and K_Hep values of ICG R₁₅≤20% and ICG R₁₅＞20% groups.

A, E, I: F0, F2, F4 stage HF pathology map (×100); B, F, J: T1native pseudo-color image, T1native measurement value is 582.31, 651.28, 700.47 ms. C, G, K: T110 min pseudo-color image, the measured value of T110 min is 311.22, 358.43, and 437.19 ms. D, H, L: T120 min pseudo-color image, T120 min measured value is 298.34, 347.56, 415.27 ms.

Boxplot of T1native (A), T110 min (B), T120 min (C) and ECV20 min (D) values with HF ranging from F0 to F4.

Fig 3. The Gadoxetate uptake model fitted to hepatic Gadoxetate concentration data (calculated from DCE-MRI images) for the four patients that underwent both pre- and post resective surgery MRI examinations. The model simulation are the green line (pre-surgery) and the orange line (post-surgery). The datapoints are green (pre-surgery) and orange (post-surgery). The column to the far right is the model fit to the mean spleen data for each of the four patients.

Fig 1. a) The mechanstic Gadoxetate uptake model. Shown here, is a schematic over the model structure. The k_in and k_eff model parameters represent the hepatic influx and efflux and are highlighted in orange. b) The study setup. Showing the time-points for the pre- and post resective surgery MRI examinations. c) Showing serum level blirubin for three of the patients that underwend both the pre- and post resective surgery MRI examinations.