Clinical hepatic diffusion weighted imaging (DWI) generally relies on mono-exponential diffusion. The aim was to demonstrate that mono-exponential diffusion in the liver is contaminated by microperfusion and that the bi-exponential model is required.
Methods
Nineteen fasting healthy volunteers were examined with DWI (seven b-values) using fat suppression and respiratory triggering (1.5 T). Five different regions in the liver were analysed regarding the mono-exponentially fitted apparent diffusion coefficient (ADC), and the bi-exponential model: molecular diffusion (Dslow), microperfusion (Dfast) and the respective fractions (fslow/fast). Data were compared using ANOVA and Kruskal–Wallis tests. Simulations were performed by repeating our data analyses, using just the DWI series acquired with b-values approximating those of previous studies.
Results
Median mono-exponentially fitted ADCs varied significantly (P < 0.001) between 1.107 and 1.423 × 10−3 mm2/s for the five regions. Bi-exponential fitted Dslow varied between 0.923 and 1.062 × 10−3 mm2/s without significant differences (P = 0.140). Dfast varied significantly, between 17.8 and 46.8 × 10−3 mm2/s (P < 0.001). F-tests showed that the diffusion data fitted the bi-exponential model significantly better than the mono-exponential model (F > 21.4, P < 0.010). These results were confirmed by the simulations.
Conclusion
ADCs of normal liver tissue are significantly dependent on the measurement location because of substantial microperfusion contamination; therefore the bi-exponential model should be used.
Key Points
Diffusion weighted MR imaging helps clinicians to differentiate tumours by diffusion properties
Fast moving water molecules experience microperfusion, slow molecules diffusion
Hepatic diffusion should be measured by bi-exponential models to avoid microperfusion contamination
Mono-exponential models are contaminated with microperfusion, resulting in apparent regional diffusion differences
Bi-exponential models are necessary to measure diffusion and microperfusion in the liver
Introduction
Diffusion weighted imaging (DWI) started with the introduction of intravoxel incoherent motion (IVIM) imaging by Le Bihan et al. in the late 1980s [1]. IVIM was developed to quantify microscopic translational motions in a voxel by defining the apparent diffusion coefficient (ADC), which integrates the effects of both molecular diffusion and microperfusion of blood in the capillary network [2]. The ADC is sensitive to all intravoxel incoherent motions and equals the diffusion coefficient Dslow if only molecular diffusion is present. However, ADCs of brain tissue were often higher than expected because of microperfusion contamination. Therefore, the IVIM theory was extended to a bi-exponential model that was used to obtain pure and separate images of molecular diffusion Dslow and microperfusion Dfast [3]. DWI was introduced in the abdominal organs from the early 1990s [4]; before that, it was primarily limited to the brain [1‐3, 5]. Subsequently, DWI of liver parenchyma and liver abnormalities was performed, however mainly by application of ADC quantification using the mono-exponential model [6‐8]. Yamada et al. then demonstrated that the effect of microperfusion significantly contributes to the ADCs of abdominal organs and hepatic lesions [9]. They concluded by application of the bi-exponential model that the molecular diffusion coefficient Dslow and microperfusion fraction f are useful parameters for the characterisation of hepatic lesions. This was supported by later studies that showed the additional value of using the bi-exponential model for the clinical evaluation of hepatic parenchyma and hepatic lesions [10‐14].
However, we noticed that most liver studies up to now rely on the ADC as a measure for molecular diffusion without taking into account microperfusion contamination [15‐17]. Only a few liver studies avoided microperfusion contamination by choosing only higher b-values starting from 50 mm2/s for the calculation of ADCs [18‐23]. Others showed that besides the MRI technique and field strength, the location of the measurement can influence the ADC significantly [21, 24‐26]. We, however, suspected that the variation of ADCs in different regions of the liver was due to microperfusion contamination. Thus, the purpose of this study was to demonstrate that mono-exponential diffusion in the liver is contaminated by microperfusion and that the bi-exponential model is required.
Anzeige
Methods and materials
Study population
The protocol of the study was approved by the hospital’s institutional review board, and informed consent was obtained from all subjects. The study population comprised 10 men and 9 women (n = 19) ranging from 20 to 62 years old (mean 32.9 years). All subjects were healthy volunteers, without relevant medical history, with a body mass index (BMI) ranging from 20 to 32 kg/m2. The only preparation before the examination was an 8-h fasting period.
MR protocols
All subjects were prospectively examined using MR imaging at 1.5 T (Magnetom Avanto, Siemens Medical Solutions, Erlangen, Germany). The body coil served as a transmitter and a six element spine matrix coil in combination with the body matrix as a receiver. After routine localiser- and T2-weighted imaging a series (b = 0, 50, 100, 250, 500, 750, 1000 s/mm2) of isotropic diffusion weighted images (DWI) were acquired using a spin echo based echo-planar imaging (EPI) sequence in combination with spectral adiabatic inversion recovery (SPAIR) fat suppression. The acquisition was gated using PACE respiratory triggering (TR = 3100–6500 ms) and was tuned with the following parameters: TE 75 ms; slice-thickness 6 mm; slice-gap 18 mm; FOV 379 × 284 mm; matrix 192 × 144; bandwidth 1735 Hz/pixel; averages 2 and parallel acquisition technique GRAPPA with acceleration factor 2. Diffusion gradients (25 mT/m) were applied in the phase, read, and z-directions separately. In total 9 transverse slices were acquired with an 18 mm slice gap to cover the whole liver within an average total acquisition time of 8.1 min (range; 4.7–11.1 min). The image acquisition took place in an interleaved mode; first slices 1, 4 and 7 were consecutively acquired with b = 0 value, then the same slices with b = 50 value and so on up to b = 1000. Subsequently, slices 2, 5 and 8 were acquired in the same way, and finally slices 3, 6 and 9 also.
Image analysis
A programmable graphical and calculus environment was used (Matlab, The Mathworks, Natick, MA, USA) to develop a tool for analysing the DWI series using mono- and bi-exponential fitting procedures. First, the DWI data were loaded and the five liver regions were defined for all 19 datasets by drawing circular regions-of-interest (ROIs) with a diameter of 22.3 mm on selected transverse 6-mm slices (Fig. 1). For each dataset three slices at three different levels were selected; medial at the level of the right portal vein, cranial above the level of the left portal vein and caudal at the level of the splenic vein. The first three regions were drawn in the medial liver slice: central in the right lobe (ROI 1, segments 5/8), dorso-lateral in the right lobe (ROI 2, segments 6/7) and ventral in the left lobe (ROI 3, segments 2/3). The fourth region was drawn in the cranial liver slice, dorso-lateral in the right lobe (ROI 4, segment 7), and the last region in the caudal liver slice, near the gallbladder (ROI 5, segment 5). Any nearby visible vascular and biliary structures were avoided. For each ROI and all seven b-values the average signal intensity S was calculated and stored. Then, the signal intensities S and all seven b-values were fitted to the bi-exponential model [3]:
where S0 is the maximum signal intensity, Dfast is the fast component representing microperfusion, ffast is the fraction of microperfusion, Dslow is the slow component representing molecular diffusion and fslow is the fraction of molecular diffusion (fslow= 1 - ffast). Equation 1 was fitted by the Nelder-Mead simplex direct search method with bound constraints (Table 1), which performs a constrained non-linear minimisation of the sum of the squared residuals [27, 28]. Finally, the conventional ADC was determined using linear regression analysis of the natural log of signal intensities S versus all seven b-values [2]:
Table 1
Fitting parameters
Bi-exponential fit (p = 3)a
Lower-bound
Higher-bound
Initial guess
ffast (unitless)
1e-8
1
0.4
Dfast (×10−3 mm2/s)
1e-5
100
10
Dslow (×10−3 mm2/s)
1e-5
10
1
aNumber of parameters
$$ S = A\exp ( - b \cdot ADC) $$
(2)
where A is the intercept of the signal.
×
Parametric maps
For one subject, parametric maps of ADC and Dslow were calculated by fitting the diffusion signal of a 4 by 4 pixels area to the mono- and bi-exponential model. The area was then moved 4 pixels until the whole liver matrix (192 × 144) was covered, resulting in a parametric map of 48 × 36. Subsequently, the parametric map was rescaled to the original matrix size by linear interpolation.
Anzeige
Influence of blood vessels
To illustrate the effects of microperfusion on the mono- and bi-exponential models, an ROI was drawn in one subject and blood vessels were included in the ROI. Then the same ROI was moved slightly until the blood vessels were excluded from the ROI. In both cases ADC, Dslow, Dfast and f were calculated and the curves presented in a figure for comparison.
Simulation study
A systematic literature search was performed in order to compare our results with previous findings. We included multi-region DWI liver studies that determined the ADC for healthy liver tissue for at least 20 patients and published a P-value for the significance of the ADC difference among the regions. Then the results of the reviewed studies were simulated by performing mono- and bi-exponential analyses on our data using the b-values and regions mentioned in the review studies. The ADC, Dslow, Dfast, fslow and ffast resulting from the simulation were then compared with those of the reviewed studies.
Statistical analysis
Statistical analysis was performed using SPSS (SPSS 18, Chicago, IL, USA). All data were tested for normality using Shapiro–Wilk tests. For normally distributed data (ADC and Dslow) one-way ANOVA tests were used to compare measurements between different liver regions. For non-normally distributed data (Dfast and f) the differences between liver regions were determined using non-parametric Kruskal–Wallis tests. The central tendency and variability of the data were quantified by the median \( {\mu_{{1/2}}} \) and the median deviation \( {\sigma_{{1/2}}} \) [29] where
For each individual fitting procedure the coefficient of correlation (goodness-of-fit) R2 was adjusted for the number of parameters p and the number of data points n:
where n = 7 for seven b-values and p = 2 or 3 for mono- and bi-exponential models respectively. F-tests were performed to evaluate whether the bi-exponential model was justified over a mono-exponential model:
$$ F = \frac{{(SS{E_{{mono}}} - SS{E_{{bi}}})/(D{F_{{mono}}} - D{F_{{bi}}})}}{{SS{E_{{bi}}}/D{F_{{bi}}}}} $$
(5)
where SSEmono and SSEbi are the sum-of-squared errors (SSE) of the mono- and bi-exponential fits respectively, and DFmono and DFbi are the degrees of freedom of both analyses (DF = n – p). All statistical analyses had a significance level of 5%.
Results
The mono-exponentially fitted ADCs were significantly different (P < 0.001) among the five regions and normally distributed (Tables 2 and 3). The left lobe showed the highest ADC (1.423 × 10−3 mm2/s) and the lowest ADC was found centrally in the medial right lobe (1.107 × 10−3 mm2/s).
Table 2
Tests of normality
Region
Segment
ADC
Dslow
Dfast
fslow/fast
1
5/8
0.798
0.929
0.022*
0.761
2
6/7
0.895
0.708
0.086
0.005*
3
2/3
0.814
0.309
0.007*
0.900
4
7
0.208
0.523
0.563
0.023*
5
5
0.230
0.735
0.167
0.663
P values (Shapiro–Wilk tests) indicate deviations from normal distribution (P < 0.05). Measurements of Dfast and f fractions were non-normally distributed, Dslow values were normally distributed
Table 3
Mono- and bi-exponential fitting results for five regions in the liver
Mono-exponential fit
Bi-exponential fit
Radj2 ≥ 0.479
Radj2 ≥ 0.969
Region
Segment
ADC (×10−3 mm2/s)
ffast (%)
Dfast (×10−3 mm2/s)
fslow (%)
Dslow (×10−3 mm2/s)
1
5/8
1.107 ± 0.101
29 ± 5
33.5 ± 12.1
71 ± 5
0.923 ± 0.148
2
6/7
1.204 ± 0.055
24 ± 3
37.8 ± 11.0
76 ± 3
1.038 ± 0.052
3
2/3
1.423 ± 0.118
47 ± 9
17.8 ± 7.1
53 ± 9
0.900 ± 0.228
4
7
1.239 ± 0.090
24 ± 5
46.8 ± 19.8
76 ± 5
1.062 ± 0.118
5
5
1.107 ± 0.102
29 ± 4
43.6 ± 9.7
71 ± 4
0.954 ± 0.120
P-value
<0.001*
<0.001†
<0.001†
<0.001†
0.140*
Data are medians ± median absolute deviations. Both mono- and bi-exponential fits were determined with seven b-values (0, 50, 100, 250, 500, 750 and 1000). Radj2 is the goodness-of-fit correlation coefficient. P values indicate the significance of differences between different regions in the liver. *One-way ANOVA. †Non-parametric Kruskal–Wallis
The bi-exponentially fitted Dslow was not shown to be significantly different (P = 0.140) among the five regions and were normally distributed (Tables 2 and 3). The median of the Dslow components varied between 0.923 and 1.062 × 10−3 mm2/s for each of the five regions.
The bi-exponentially fitted Dfast was significantly different (P < 0.001) between the five regions and non-normally distributed (Tables 2 and 3). The median of the Dfast components varied between 17.8 and 46.8 × 10−3 mm2/s for the five regions.
The left lobe showed the highest fraction of microperfusion ffast (47%) and the lowest fraction of diffusion fslow (53%). The dorso-lateral right lobe showed the lowest fraction of microperfusion ffast (24%) and the highest fraction of molecular diffusion fslow (76%). The fslow and ffast fractions were significantly different among the five regions (P < 0.001) (Table 3).
Anzeige
The adjusted R2 showed that the bi-exponential model fitted better to the diffusion data than the mono-exponential model for each liver region (Radj, bi2 > Radj, mono2). Furthermore, the F-tests showed that the diffusion data fitted significantly better to the bi-exponential model (F > 21.4, P < 0.010) than to the mono-exponential model in all individual fitting procedures.
Parametric maps
Diffusion weighted imaging data of the medial liver of one subject were used to calculate ADC and Dslow maps (Fig. 2). The Dslow map showed a more homogeneous distribution of the bi-exponential fitted slow diffusion throughout the liver. The ADC map showed more differing values between different regions of the liver, especially near blood vessels and in the left lobe.
×
Influence of blood vessels
In one subject, the effects of microperfusion on the mono- and bi-exponential models were illustrated by including blood vessels in the ROI; this showed a large influence of microperfusion on the ADC (Fig. 3). When the blood vessels were excluded from the ROI a decreased ADC was observed. However, Dslow was similar to when blood vessels were included in the ROI.
×
Simulation study
The systematic literature search resulted in six multi-region DWI studies of the liver in which a minimal of 20 patients were included and in which between two and four different regions of healthy liver parenchyma were analysed [13, 19, 21, 25, 26, 30]. Studies in which mono-exponential analyses were performed showed ADCs between 1.00 and 2.69 × 10−3 mm2/s using sets of two, three or five b-value combinations between 0 and 1300 s/mm2 (Table 4). We found ADCs between 0.94 and 2.39 × 10−3 mm2/s when using the same b-value combinations. All mono-exponential studies reported significantly (P < 0.05) different ADCs between the different liver regions, this was confirmed by our simulation study (P < 0.001).
Table 4
Reviewed data simulated with mono-exponential analysis
ADCs are means in units of 10−3 mm2/s. P values (one-way ANOVA) indicate the significance of differences between different regions in the liver. †P value was not mentioned; however the significance level was 5%. aMürtz et al. [19] did not evaluate regions in the left lobe; therefore we excluded ROI 3 (left lobe) from the simulation
Anzeige
In one study a bi-exponential analysis of two different regions of liver parenchyma was performed using a series of ten b-values between 0 and 800 s/mm2 (Table 5) [13]. They did not find significantly different Dslow components among different liver regions (P = 0.5); this was confirmed by our simulation study (P = 0.105).
Table 5
Reviewed data simulated with bi-exponential analysis
Data are means. P values indicate the significance of differences between different regions in the liver. aOne-way ANOVA. bNon-parametric Kruskal–Wallis
Discussion
The aim of this study was to demonstrate that mono-exponential diffusion in the liver is contaminated by microperfusion. We suspected that variation of ADCs in different regions of the liver is due to microperfusion contamination and that therefore the bi-exponential model is required. We found that the ADC of normal liver tissue is substantially dependent on the location; however, the bi-exponentially fitted Dslow is not dependent on the location of the measurement. We used terminology according to the suggestions of Guiu and Cercueil who convincingly advocated for the use of Dslow to describe molecular diffusion properties and Dfast to describe microperfusion [31].
Our results agree with earlier studies reporting significant differences among ADCs obtained in different regions in the liver [19, 21, 25, 26, 30]. ADCs in our study (1.107 to 1.423 × 10−3 mm2/s) were consistent with those of previous reports, which showed a large range (0.69 to 2.69 × 10−3 mm2/s) of ADCs for normal liver tissue [15, 25, 32]. The results on Dslow were supported by Luciani et al.; they also found no significant differences between Dslow of normal liver tissue [13]. They found that Dslow varied between 1.02 and 1.16 × 10−3 mm2/s; this is comparable to our range (0.98 to 1.18 × 10−3 mm2/s) when we simulated their results using our data and similar b-values.
The demonstrated regional dependence of the ADC contrary to the non-regional dependence of Dslow can be explained from the differences between the mono- and the bi-exponential models. The mono-exponentially fitted ADC is to a high degree sensitive to microperfusion, which was already shown by Le Bihan et al. in the brain [2]. When the DWI sequence contains b-values in the microperfusion range, and the microperfusion is relatively high compared with the diffusion component, then the model will result in a high ADC compared with Dslow. When there is no microperfusion, which has been reported for fibroglandular breast tissue, the ADC will be comparable to Dslow [33]. Hence the ADC of the liver is to a large extent dependent on microperfusion, especially when the DWI sequence contains several b-values in the microperfusion range [14]. This is why in some studies researchers tried to choose b-values not within the microperfusion range [18‐23]. For example, Perman et al. observed that the ADC value for liver decreased from 1.36 to 0.98 × 10−3 mm2/s when the b = 0 value was omitted; this is similar to the range of Dslow that we found. In a recent consensus report on DWI, it was recommended to use two b-values (>100 and between 500 and 1000 mm2/s) for ADC assessments [34]. It is, however, difficult to choose b-values such that microperfusion contamination of the ADC is avoided, because the microperfusion effects of the tissue are not known beforehand. We demonstrated in our simulation study that, although low b-values were excluded, the ADC of liver tissue was to a large extent contaminated by microperfusion and that this resulted in apparent differences of the diffusion between liver regions. The simulation study also showed that the variety of ADC values of healthy liver tissue published in the literature, were probably caused by the choice of b-values, and the mono-exponential model itself. This is a pitfall of using the ADC as a measure for molecular diffusion; there is no optimal combination of b-values, because the amount of microperfusion determines the optimal sequence of b-values and this is not known a priori. Recently, a similar simulation study was performed in the kidneys; they also found that the variability of the ADC in the kidneys is caused by the use of the mono-exponential model [35].
Anzeige
The microperfusion component Dfast of the bi-exponential model significantly depends on the location of the measurement. This contradicts with the results of Luciani et al., who did not find a location dependency of Dfast, possibly because they compared just two regions (left and right lobe) [13]. The range (71.0–85.1 × 10−3 mm2/s) was higher than what we found (27.9–57.2 × 10−3 mm2/s) when we simulated their results using our data and similar b-values. This may reflect the use of different calculation methods. In our study f, Dfast and Dslow of an ROI were calculated by taking the medians of the underlying b-maps as an input for the fitting procedure, where Luciani et al. first calculated Dfast, Dslow and f maps on a pixel-by-pixel based fitting procedure, after which the average of an ROI was calculated. In addition, they used six b-values under 100 s/mm2 (versus two in our study), which tends to decrease the uncertainty of the fitting algorithm, and increase the slope of the curve close to zero. This might explain the higher Dfast and the decreased range of Dfast in their study. However, we have shown that even with seven (rather than 16 b-values) the bi-exponential fits are already extremely accurate, which is supported by the nearly identical values of Dslow in their study and our simulation study. Too little b-values under b = 100 s/mm2 can however hamper an accurate determination of Dfast, especially when analysing tissues with high Dfast values, which can be expected in pathology or near blood vessels.
The increase in the ADC in the left lobe, which was demonstrated in Fig. 2, is usually explained from the increased cardiac motion in the left lobe [19, 25, 36‐38]. However, we found that the fraction of microperfusion ffast in the left lobe was almost a factor two higher than in other liver locations. Although pseudo-anisotropy is a known artefact of respiratory triggering in the liver [39], we suspect that the increased ADC in the left lobe may be caused by extensive microperfusion contamination of the ADC and to a lesser extent by either cardiac or respiratory artefacts. This is supported by a study on the hepatic perfusion of eight hepatic segments by dual-source computed tomography [40]. They found that the hepatic perfusion index (HPI) was significantly higher in segment 3 (left lobe) than in segments 5 to 8 (right lobe), and suggested that this might be related to the anatomy of the liver vessels. Another study using pharmacokinetic analysis of dynamic contrast-enhanced MRI demonstrated that regional variations in liver microcirculation can be displayed by colour-coded parameter maps [41]. They found a minor variation of perfusion in an apical section of a transplanted liver. The left part of the liver, corresponding to segment 2, showed a different perfusion rate than the right part of the liver. In an experimental study on rats using perfusion CT, the relative blood flow in the left lobe was 17% higher than in the right lobe of the liver [42]. We suspect that these regional variations in the density of small blood vessels and capillaries caused the heterogeneous appearance of microperfusion throughout the liver. This is supported by our analysis of including a blood vessel in the ROI, which resulted in more microperfusion contamination than when blood vessels were avoided.
Some studies have reported on age-related changes in liver structure and function [43, 44]. However, Pasquinelli et al. showed no significant variations in liver DWI quantitative parameters according to the age of the subject [45]. Therefore, we assumed that the possible effects of age in our cohort are far smaller than the demonstrated significant region dependency of the ADC in the liver. Although the conclusions in this study are drawn from healthy volunteers, we suspect that the effects of microperfusion are much larger in pathology, and should therefore be applicable to patient data also.
In conclusion, the ADC of normal liver tissue is significantly dependent on the measurement location because of substantial microperfusion contamination; therefore the bi-exponential model should be used. Currently, the diagnostic use of DWI for discriminating hepatic masses, liver cirrhosis and fibrosis is mainly based on the ADC, and the discrimination between diseased and healthy tissue may therefore be hampered [6‐8, 15‐18, 20, 21, 23, 30, 37, 46]. Thus, the bi-exponential model is essential for the future development of the clinical diagnostic application of DWI in the liver.
Open Access
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
Open AccessThis is an open access article distributed under the terms of the Creative Commons Attribution Noncommercial License (https://creativecommons.org/licenses/by-nc/2.0), which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
Mit e.Med Radiologie erhalten Sie Zugang zu CME-Fortbildungen des Fachgebietes Radiologie, den Premium-Inhalten der radiologischen Fachzeitschriften, inklusive einer gedruckten Radiologie-Zeitschrift Ihrer Wahl.
Akuter Schwindel stellt oft eine diagnostische Herausforderung dar. Wie nützlich dabei eine MRT ist, hat eine Studie aus Finnland untersucht. Immerhin einer von sechs Patienten wurde mit akutem ischämischem Schlaganfall diagnostiziert.
Routinemäßige Mammografien helfen, Brustkrebs frühzeitig zu erkennen. Anhand der Röntgenuntersuchung lassen sich aber auch kardiovaskuläre Risikopatientinnen identifizieren. Als zuverlässiger Anhaltspunkt gilt die Verkalkung der Brustarterien.
Die Empfehlungen zur Therapie des Pankreaskarzinoms wurden um zwei Off-Label-Anwendungen erweitert. Und auch im Bereich der Früherkennung gibt es Aktualisierungen.
Im Choosing-Wisely-Programm, das für die deutsche Initiative „Klug entscheiden“ Pate gestanden hat, sind erstmals Empfehlungen zum Umgang mit Notfällen von Kindern erschienen. Fünf Dinge gilt es demnach zu vermeiden.
Update Radiologie
Bestellen Sie unseren Fach-Newsletterund bleiben Sie gut informiert.