To determine whether diffusion-weighted echo-planar (EP) MR images with very small, small, and large gradient b-factors are useful in evaluating hepatic lesions and hepatic parenchyma.
To determine whether diffusion-weighted echo-planar (EP) MR images with very small, small, and large gradient b-factors are useful in evaluating hepatic lesions and hepatic parenchyma.
Approximate values of the apparent diffusion coefficients for diffusion (D) and for flowing spins (D*) for 96 hepatic lesions (26 hepatocellular carcinomas [HCCs], 28 metastases, 26 hemangiomas, and 16 cysts) and the non-lesion-bearing regions of parenchyma in 78 livers (50 noncirrhotic and 28 cirrhotic) were calculated from EP images (modified for gradient b-factors of 3, 50, and 300 second/mm2).
Liver cysts and noncirrhotic livers showed statistically higher mean D* values than HCCs, hemangiomas, metastases, and cirrhotic livers (P < 0.05 on Scheffé post hoc analysis). Liver cysts showed statistically higher mean D values than HCCs, metastases, noncirrhotic livers, and cirrhotic livers (P < 0.05). Liver hemangiomas showed statistically higher mean D values than HCCs, noncirrhotic livers, and cirrhotic livers (P < 0.05).
The D* value in addition to the D value may be useful for evaluating the nature of diffusion and flowing spins in hepatic lesions and hepatic parenchyma. J. Magn. Reson. Imaging 2006. © 2006 Wiley-Liss, Inc.
THE ECHO-PLANAR (EP) IMAGING TECHNIQUE was developed to provide very high speed imaging capability, because the entire set of echoes necessary to form an image can be collected within a single acquisition period. It became important in the examination of organs in which artifacts due to gross physiologic motion often degrade MR images (1). The use of EP imaging for abdominal imaging was introduced by Stehling et al  (2). The EP imaging with a preparation pulse sensitive to diffusion (diffusion-weighted EP imaging) allows diffusion-weighted images to be obtained within a single breath-hold. The most useful application of this imaging technique is in the early stage detection of brain ischemia (3). Recently, abdominal diffusion studies using single-shot EP imaging have been performed in vivo. There have also been reports that EP imaging with small and large gradient b-factors used for hepatic lesion evaluation might be valuable in characterizing focal liver lesions (4, 5).
The term intravoxel incoherent motion is used in reference to the microscopic translations that occur in voxels on MR images (6). Intravoxel incoherent motions include not only the molecular diffusion of water, but also flowing spins such as perfusion since perfusion can be considered an incoherent motion when blood circulates in a pseudorandomly organized capillary network at the voxel level. Perfusion effects usually cause larger signal attenuation than diffusion effects on images with a small gradient b-factor (7). On the basis of this theory, the apparent diffusion coefficient (ADC) value calculated from images with very small and small gradient b-factors is thought to be the value most strongly influenced by flowing spins rather than molecular diffusion. The diffusion value can be approximated from images with small and large gradient b-factors. In the present study, the isolated ADC values in relation to flowing spins and diffusion were approximately calculated for hepatic lesions and hepatic parenchyma using EP images with very small, small, and large gradient b-factors. Subsequently, we determined whether these values were useful in evaluating hepatic lesions and hepatic parenchyma. In addition, we performed phantom studies (simulating liver perfusion and simulating liver cyst scanned after breath movement) to evaluate whether these approximated values could reasonably reflect diffusion- and flowing spins-related effects, although it is not strictly correct to describe the diffusion- and flowing spins-related effects using diffusion-weighted EP imaging with only three values of gradient b-factor.
The subjects consisted of 87 patients with hepatic lesions that had been detected on ultrasound and/or computed tomography (CT), and all were referred for evaluation of liver lesions with MR imaging. Informed consent was obtained from all patients. EP protocols could not be performed or were disregarded for four patients who were incapable of holding their breath. In five patients having lesions less than 2 cm in diameter, EP protocols were not performed because the partial volume effects would likely contaminate the and diffusion- flowing spin-related values for such small lesions. The remaining 78 patients (43 male, 35 female; age range, 36–91 years; mean age, 65 years) were examined with MR imaging including EP sequences. In patients with a single hepatic lesion, the lesion and one region of hepatic parenchyma were evaluated. In patients with multiple lesions, two lesions and one region of hepatic parenchyma were evaluated. As a consequence, a total of 96 hepatic lesions and 78 regions of hepatic parenchyma were evaluated. Fifty-four lesions were malignant tumors, 26 hepatocellular carcinomas (HCCs), and 28 metastases. Of these, 10 HCCs and 13 metastases were diagnosed at biopsy or operation. The diagnosis of the other 16 HCCs was based on the appearance of nodular enhancement immediately after injection of a bolus of iohexol (Omnipaque 300; Daiichi Seiyaku, Tokyo, Japan) showing decreased enhancement on delayed images obtained at dynamic contrast-enhanced CT, or when hepatic lesions larger than 2 cm on CT and ultrasonography were found in patients with alpha-fetoprotein levels more than 200 g/m3, or with PIVKA-II (protein induced by vitamin K absence or antagonist-II) more than 100 kAU/m3. In the remaining 15 metastases, lesion progression was diagnosed by serial CT and ultrasound examination. The sites of primary disease included the colon (N = 8), stomach (N = 7), lung (N = 5), rectum (N = 4), breast (N = 3), and ovary (N = 1). The diagnosis of 26 hemangiomas was established with the appearance of slightly irregular or nodular peripheral enhancement immediately after injection of a bolus of contrast medium that then gradually filled in toward the center on delayed images obtained from dynamic contrast-enhanced CT. The diagnosis of 16 cysts was established by the absence of enhancing components in the lesion when precontrast and postcontrast CT images were compared. Twenty-eight patients had underlying cirrhosis, which was proven from biopsy, surgery, or by clinical data.
MR imaging was performed with a 1.5-T MRI scanner with self-shielded gradients (Signa Horizon; GE, Milwaukee, WI, USA). Conventional T1-weighted (TR/TE = 500/14 msec) spin-echo and T2-weighted (echo train length = 8, TR/TE = 5000/95–105 msec) fast spin-echo images (field of view [FOV] = 32 × 24 cm, matrix size = 512 × 224, slice thickness = 7 mm, acquisitions equals two) were also obtained in all patients. In 42 patients, T1-weighted spin-echo MR images (TR/TE = 500/14 msec) were also obtained following the intravenous administration of 0.1 mmol/kg gadopentetate dimeglumine (Magnevist, Schering AG, Berlin, Germany). In 15 patients with liver metastases, T2-weighted (echo train length = 8, TR/TE = 5000/95–105 msec) fast spin-echo images were also obtained 1.5–4 hours after completion of a slow intravenous infusion of 10 μmol Fe/kg ferumoxides (Feridex, Tanabe Seiyaku, Osaka, Japan) diluted in 100 cm3 of 0.05 g/cm3 glucose solution and infused at a rate of 3 cm3/minute.
In all patients, multislice EP scans were performed using a spin-echo type of preparation (94.9 msec TE, 12 slice levels, 7 mm slice thickness, and 3 mm gap). EP scans were always performed before constant agent administration. The first scan was a T2-weighted image without diffusion gradients. The second and third scans were T2-weighted images modified by adding smaller and larger diffusion gradients on both sides of the 180° pulse along frequency encoding directions, respectively. Total scan time for each multislice EP sequence was two seconds. Diffusion-weighted EP sequences were performed under breath-hold to diminish, as much as possible, slice-level differences of the liver among diffusion-weighted EP sequences with three values of gradient b-factor. The gradient b-factor was determined by the time integral of the diffusion gradients, imaging gradients, and cross-terms between the imaging gradients and the diffusion gradients. The general expression of the gradient b-factor (b) is given by the following equation (8):
where γ is the gyromagnetic ratio (42.576 MHz/T), (t) is the instantaneous gradient vector at t, and (t) is −(t) for t > (TE)/2. The gradient b-factors of the EP sequences (b0, b1, and b2) were set at 3, 50, and 300 second/mm2, respectively. The preparation period was followed by an EP readout of 70 msec. Fat suppression (chemical shift saturation method) was introduced to remove chemical shift artifacts, and 128 lines of k-space data were acquired using the half-Fourier technique. The other parameters were a FOV of 35 × 25 cm and a 128 × 128 matrix size. Patients were instructed to take a deep breath and hold it during scanning for each EP sequence.
The signal intensity was measured in the hepatic lesions and hepatic parenchyma for obtained EP images. A circular region of interest (ROI) was located in the center of liver cysts and hemangiomas, with the outer margin of the ROI being at least 5 mm away from the outer margin of the hepatic lesion (0.9–12.3 cm2). In HCCs and metastases, ROIs (0.3–9.3 cm2) were located in the enhanced solid component whenever possible using the set of T1-weighted images or CT images with and without enhancement. Small ROIs (less than 0.9 cm2) were used for tumors with central necrosis (one HCC and six metastases). Another circular ROI (2.3–15.4 cm2) was located in the central portion of the right hepatic lobe in order to prevent the inclusion of large intrahepatic vessels, other hepatic lesions, artifacts, and extrahepatic regions.
As in the diffusion-weighted spin-echo technique (3), the ADC value of the EP images with different gradient b-factors was calculated according to the equation:
where S′ and S″ represent the signal intensity of the images with different gradient b-factors, and b̄ is the difference between gradient b-factors. We used a simple approach to separate the effects of diffusion and flowing spins. When S0, S1, and S2 were defined as the signal intensity of the images with very small (b0), small (b1), and large (b2) gradient b-factors, the D (ADC for diffusion) value calculated from Eq.  using b1, b2, S1, and S2 is more affected by diffusion than flowing spins (7). The f value (the deviation factor representing the fractional volume occupied in the voxel by flowing spins) can be obtained when the contribution of the flowing spins to signal intensity is almost negligible in S1 and S2:
where Sa is the expected signal intensity excluding the flowing component at b = b0, and ln(Sa) can be obtained from the following equation:
The signal difference between S0 and Sa is almost exclusively caused by flowing spins (excluding signal changes in relation to the value D), and D* (ADC for flowing spins) value can be approximated as follows:
To determine the statistical significance for differences in calculated D and D* values among groups and between the groups, a series of Kruskal-Wallis tests and Scheffé post hoc analyses at 0.05 level were conducted.
D and D* values were measured using a phantom made of a series of 4-cm diameter tubes containing water or acetone at 25°C. Using the same parameter to those in the in vivo study (gradient b-factors of 3, 50, and 300 second/mm2), EP sequences were performed.
To confirm that the EP images with three values of b-factor (3, 50, and 300 second/mm2) used in our in vivo study were valid for evaluation of liver perfusion and diffusion, an in vitro study using a perfusion phantom was also performed at 25°C. Both perfusion and static phantoms consisted of syringes (3.4 cm for the diameter of the lumen) filled with a dextran gel (Sephadex G-25 Coarse; Amersham Bioscience, Uppsala, Sweden) and water. The individual beads of gel, which contain many reentrant cavities, show a somewhat larger mean wet diameter (0.32 mm) compared to the mean length of axial sinusoids (0.25 mm) radially running in a liver lobule (9, 10). The Sephadex has an approximately 30% volume fraction of flowing fluid (or extra bead volume), which simulates the vascular volume of the liver having 25 to 30 cm3 of blood per 100 g (11). Slice direction was orthogonal to the overall flow direction. Average velocity through the Sephadex (vavg) in the flowing phantom was calculated as:
The vavg of 0.3 mm/second was used for the perfusion phantom to simulate sinusoidal blood flow velocity (12). The same EP sequences with the same parameters as the in vivo study except for much further multiple (35 points) gradient b-factors (3–1000 second/mm2) were performed eight times, and a round ROI (2.7 cm2) was placed in the center of each phantom.
To evaluate whether liver cyst images could be affected by flowing-spin effects in relation to diaphragmatic or liver motions before breath-hold EP scans, we performed the following in vitro studies. To simulate these situations, we made a phantom consisted of a not fully inflated artificial cyst (made from polyurethane condom [Sagami Original 0.03 mm L-size, Sagami Rubber Ind. Co. Ltd., Kanagawa, Japan], inflated with 30 mL of normal saline solution) and 1% agar gel (having almost identical elasticity to one of a living liver) (13, 14) in which the artificial cyst was enclosed. The phantom simulating liver cyst was enveloped by a thin polyethylene bag (0.01 mm in thickness) and sunk in a cylindroid tank filled with water at 25°C. The phantom was set to be actuated with periodic piston movement. We performed each EP scan within the following two conditions: 1) three seconds after the phantom that had been actuated with periodic piston movement (5.6 cm of amplitude, and 0.3 cycles per second) stopped, simulating liver movement at deep breathing (15) and successive breath-hold scan; and 2) during the constantly static state of the phantom without piston movement before the scan. The same EP sequences with the same parameters as the perfusion phantom study, using multiple (35 points) gradient b-factors (3–1000 second/mm2), were performed eight times in each phantom study, and a round ROI (5.3 cm2) was placed in the center of the artificial liver cyst.
The D and D* values for the liver from one volunteer were repeatedly measured (11 times, one measurement/day), and the coefficients of variation (SD/mean value) were calculated to gain information regarding the precision of the in vivo results. Using the same parameters as the ones for the in vivo study (gradient b-factors of 3, 50, and 300 second/mm2) EP measurements were performed and a ROI (12 cm2) was placed in the center of the right hepatic lobe.
The only motion expected in the phantoms resulted from diffusion since care was taken to avoid thermal convection. The D values for water and acetone were, respectively, 2.37 and 4.61 × 10−3 mm2/second, which agreed with Le Bihan et al (6). The D* values for water and acetone were close to 0 (0.077 and 0.084 × 10−3 mm2/second).
The logarithm of the mean signal intensity for the flowing and static phantoms is depicted as a function of the gradient b-factor in Fig. 1. Inclination of the graphs (calculated from regression analyses) at a gradient b-factor revealed a corresponding ADC value. ADC values from the perfusion phantom were significantly higher than the values from the static phantom at a gradient b-factor of less than 60 second/mm2, were mildly higher at a gradient b-factor of 60–120 second/mm2, and very close at a gradient b-factor greater than 120 second/mm2.
The logarithm of the mean signal intensity for the artificial liver cyst when the cyst was static at all time (the constantly static cyst) and when the same cyst was scanned in static condition three seconds after piston movement simulating liver movement at deep breathing before breath-hold EP scans (the static cyst after piston movement) is depicted as a function of the gradient b-factor in Fig. 2. Inclination of the curve and linear fitting lines (calculated from regression analyses) at a gradient b-factor revealed a corresponding ADC value. ADC values from the static cyst after piston movement tended to be higher than the values from the static cyst for all b-factors. The difference in these ADC values was greatest for b-factors of less than 50 second/mm2, somewhat smaller for b-factors of 50–180 second/mm2, and least for b-factors of over 180 second/mm2.
The D and D* values for 96 hepatic lesions and 78 hepatic parenchyma regions were obtained (Figs. 3 and 4). Focal signal attenuation of the liver when adjacent to the heart (Fig. 5) was often demonstrated in EP images (considered to be caused by phase errors occurring during application of the diffusion gradients in relation to focal liver motions to be affected by cardiac pulsation). Susceptibility artifacts from gastrointestinal gas and air in the lungs were mild and at a negligible level for evaluation of liver and hepatic lesions. Except for the artifacts, the signal intensities of the liver and hepatic lesions were substantially preserved in comparison with the background in all evaluated portions of the EP images.
The D values (mean ± SD) of the hepatic lesions and parenchyma are shown in Table 1, and their scatter plots in Fig. 3. These groups, ranked in descending order of the mean values, were cysts, hemangiomas, metastases, cirrhotic livers, HCCs, and noncirrhotic livers. Relatively high SDs were found for metastases, hemangiomas, and cysts. The mean D values among these groups were significantly different (P < 0.05 on Kruskal-Wallis test). The mean D values for the cysts were higher than those for the other groups except for hemangiomas (P < 0.05 on Scheffé post hoc analysis). The mean D values for hemangiomas were higher than those for HCCs, cirrhotic livers, and noncirrhotic livers (P < 0.05 on Scheffé post hoc analysis). The mean D-value differences between any other groups were insignificant (P > 0.05 on Scheffé post hoc analysis).
The D* values (mean ± SD) of hepatic lesions and parenchyma are shown in Table 1, and their scatter plots in Fig. 4. These groups, in descending order of the mean value, were cysts, noncirrhotic livers, hemangiomas, cirrhotic livers, metastases, and HCCs. Relative high SDs were demonstrated in hemangiomas and cysts. The mean D* values among these groups were significantly different (P < 0.05 on Kruskal-Wallis test). The mean D* values for the cysts were higher than those for the other groups except for noncirrhotic livers (P < 0.05 on Scheffé post hoc analysis). The mean D* values for noncirrhotic livers were higher than those for the other groups except for cysts (P < 0.05 on Scheffé post hoc analysis). The mean D*-value differences between any other groups were insignificant (P > 0.05 on Scheffé post hoc analysis).
The coefficients of variation for D and D* values of the liver of a volunteer were 25.4% and 28.2%, respectively.
Macroscopically, microcirculation appears as a random walk process, similar to molecular diffusion. In an idealized model, the D* value for perfusion effects can be easily calculated (7):
where <l> is the mean capillary segment length (distance for which a capillary segment can be considered straight) and <v> is the mean velocity of flowing water molecules in the blood. Using data from the literature for <l> (mean length of axial sinusoids of liver) and <v> (mean velocity of blood flow in sinusoids) of 0.25 mm and 0.27–0.41 mm/second, respectively (9, 10), the D* value for sinusoids is expected to be approximately 11.3–17.1 × 10−3 mm2/second. This simulated value is lower than the D* value for the flowing phantom data using images with a gradient b-factor of 3, 50, and 300 second/mm2 (22.9 × 10−3 mm2/second). However, when the D* value of the flowing phantom is corrected by Eq.  using <l> of 0.25 mm instead of 0.32 mm, the D* value (17.8 x× 10−3 mm2/second) is approximately that of the simulated D* value. The simulated D* value is higher than our in vivo D* values for noncirrhotic livers (6.68 ± 3.49 × 10−3 mm2/second), probably due to the slower intermixing of flowing blood in shorter branches of sinusoids (having smaller <l> and lower <v> values in Eq. ). The flowing phantom study also demonstrates that liver perfusion effects can be detected when the gradient b-factors are set to less than 60 second/mm2. These results demonstrate that EP images with a gradient b-factor of 3, 50, and 300 second/mm2 are proper for evaluation of liver perfusion, although perfusion effects (D* value) may be subdued in some degree because the D value may be contaminated by flowing-spin effects at this setting. To our knowledge the evaluation of flowing spin contributions (such as perfusion effects) on hepatic lesions and hepatic parenchyma, using a set of MR images with such very small and small gradient b-factors (e.g., less than 60 second/mm2), has not been reported, with the exception of one report using centric-reordered turboFLASH images (16). However, turboFLASH imaging has an inherently lower signal to noise ratio than EP imaging, and the D* and D values in reproducibility tests using turboFLASH images showed a higher coefficient of variation than the values using EP images, even when each turboFLASH image was generated from 12 times scans at a single slice level (16). Furthermore, inherent T1 contamination for turboFLASH imaging might affect these values obtained, although centric reordered k-space sampling had been introduced to reduce this effect in the report (16). Therefore, we speculate that reevaluation of flowing spin contributions using EP images is meaningful.
The mean D* values for noncirrhotic livers were significantly higher than the values for HCCs, hemangiomas, and metastases. We consider that the higher mean D* values for the noncirrhotic liver than for the hepatic lesions (except for cysts) are reasonable because the liver is a vascular organ having a rich vascular bed with a dual afferent blood supply from the portal vein and hepatic artery (1603 ± 144 mL/minute of total blood flow) (17). Noncirrhotic livers also displayed statistically higher mean D* values than cirrhotic livers. This may reflect decreased vascularity (smaller <v> value in Eq. ), a decreased vascular pool, or distorted parenchymal and vascular structures (smaller <l> value in Eq.) in the cirrhotic liver (18) in relation to having increased fibrotic interstitium.
Liver cysts had the highest mean D and D* values in both the hepatic lesion and hepatic parenchyma groups. We believe that these D value results are reasonable because cysts have a watery content. The D* value results are entirely different from a previous report in which the flowing-spin effects for liver cysts were not detected (19). However, our phantom study indicates that the ADC values of liver cysts were possible to show unrealistically higher than pure water diffusion values at any gradient b-factors even if liver cysts were static during breath-hold EP scans, and that the D* values of the liver cysts were more largely affected than the D values on this condition. These results reveal it is possible that liver cysts can be highly affected by flowing-spin effects even in the absence of perfusion. In the condition that breath movement fluctuates both the liver and the coexisting cysts before the breath-hold EP scans, it is very likely that during the scans, even though the patients hold their breath and the liver cysts were static, lasting turbulent flow due to inertial forces may cause flowing-spin effects. We think that the higher SDs for the D* and D values of liver cysts might be caused by various levels of these flowing spins effects among liver cysts.
Since hemangiomas consist of a large pool of stagnant and flowing blood within a cavernous structure (1, 19), we expected that this would result in significantly higher mean D and D* values than those of some other groups. Hemangiomas had statistically higher mean D values than HCCs, cirrhotic liver, and noncirrhotic liver. However, mean D* values of hemangiomas were not significantly higher than the values of the other groups. We consider that the unpredicted D* value results may be caused by the variety of <l> and <v> values in Eq.  among hemngiomas, because hemangiomas showed larger SDs of the D* values.
The mean D value for the liver calculated from images with a gradient b-factor of 50 and 300 second/mm2 (1.16 × 10−3 mm2/second) on our setting is similar to the reported mean diffusion value for the liver (1.39 × 10−3 mm2/second) using gradient b-factors of 2–400 second/mm2 (1), and is relatively higher than another reported value (0.76 × 10−3 mm2/second) using gradient b-factors of 30–1100 second/mm2 (19). These results reveal that the D values may be contaminated by flowing-spin effects to some degree in our setting using a gradient b-factor of 50 and 300 second/mm2. The D values calculated from images with larger gradient b-factors than those used in our study (more than 300 second/mm2) are reported to be useful for differentiating between solid component and watery content (1, 4, 19). However, in our setting, differences between solid component and watery content using D values were also significant. Furthermore, statistically significant mean D* value differences were demonstrated among solid components (noncirrhotic livers showed larger mean D* values than HCCs, metastases, and cirrhotic livers), although there were still prominent D and D* values overlapping among these groups (Figs. 3 and 4). Strictly speaking, even when using images with larger gradient b-factors, it is impossible to precisely discriminate slowly moving flowing fluid, free diffusion, and restricted diffusion in tissues in which these effects, as well as fluids with various viscosities, are intricately intermingled. In addition, the images from much larger gradient b-factors have inherent drawbacks in lower signal intensity ratios and larger signal intensity errors than from images with smaller gradient b-factors. These drawbacks may cause larger errors for the measured diffusion values and falsely minimize the measured diffusion values in low intensity areas on T2-weighted images (such as the low intensity hepatic parenchyma of hemosiderosis) when the signal intensity of these areas are diminished relative to the signal intensity of the background on images with large gradient b-factors. Therefore, we believe that EP images with the gradient b-factor of 50 and 300 second/mm2 are not necessarily improper for evaluation of liver diffusion.
Our approach for evaluating hepatic lesions and hepatic parenchyma using EP imaging with very small, small, and large diffusion gradients presents several potential limitations. First, there is a need to perform the EP sequences several times (i.e., sequences with three different gradient b-factors). There can be mild slice level differences among obtained EP images, although multislice EP scans were introduced. In addition, small ROIs were used for evaluation of tumors with central necrosis. In such cases, D* and D values obtained may be contaminated by the partial volume effect of adjacent hepatic parenchyma and necrotic components. This is especially problematic when metastases for evaluation have relatively more central necrosis than viable components, and these might contribute to relatively large SDs of D and D* values of metastases. EP sequences are sensitive to both time-dependent and time-independent local field changes, or susceptibility artifacts (20). To reduce these artifacts, a half-Fourier technique to reduce the acquisition time or reduce the number of echoes was introduced to the EP scans. Although these artifacts were moderately prominent near the heart (Fig. 5) and mild from intestinal gas and lungs in the obtained EP images, hepatic lesions could be evaluated in all cases. However, we believe that these artifacts derived from EP imaging will be able to be reduced with the introduction of parallel imaging because this technique can markedly reduce data acquisition time. Hence, we expect that this technique might reduce errors in measurements, reduced D and D* value overlaps between the groups, and result in an increase in clinical utility for evaluation of D and D* values of hepatic lesions and hepatic parenchyma.
In conclusion, the flowing spin-related D* values and diffusion-related D values calculated from EP images with very small, small, and large gradient b-factors (3, 50, and 300 second/mm2) are suitable for evaluation of hepatic lesions and hepatic parenchyma.