Journal of Gastroenterology

, Volume 47, Issue 4, pp 477–485

Value of post-vascular phase (Kupffer imaging) by contrast-enhanced ultrasonography using Sonazoid in the detection of hepatocellular carcinoma


  • Eriko Goto
    • Department of Gastroenterology, Graduate School of MedicineThe University of Tokyo
  • Ryota Masuzaki
    • Department of Gastroenterology, Graduate School of MedicineThe University of Tokyo
    • Department of Gastroenterology, Graduate School of MedicineThe University of Tokyo
  • Yuji Kondo
    • Department of Gastroenterology, Graduate School of MedicineThe University of Tokyo
  • Jun Imamura
    • Department of Gastroenterology, Graduate School of MedicineThe University of Tokyo
  • Tadashi Goto
    • Department of Gastroenterology, Graduate School of MedicineThe University of Tokyo
  • Hitoshi Ikeda
    • Department of Clinical LaboratoryThe University of Tokyo Hospital
  • Masaaki Akahane
    • Department of RadiologyThe University of Tokyo
  • Shuichiro Shiina
    • Department of Gastroenterology, Graduate School of MedicineThe University of Tokyo
  • Masao Omata
    • Yamanashi Prefectural Hospital Organization
  • Haruhiko Yoshida
    • Department of Gastroenterology, Graduate School of MedicineThe University of Tokyo
  • Kazuhiko Koike
    • Department of Gastroenterology, Graduate School of MedicineThe University of Tokyo
Original Article—Liver, Pancreas, and Biliary Tract

DOI: 10.1007/s00535-011-0512-9

Cite this article as:
Goto, E., Masuzaki, R., Tateishi, R. et al. J Gastroenterol (2012) 47: 477. doi:10.1007/s00535-011-0512-9



We evaluated the sensitivity and specificity of post-vascular phase (Kupffer imaging) by contrast-enhanced ultrasonography (CEUS) using perflubutane microbubbles (Sonazoid) in comparison with conventional B-mode ultrasonography (US) for the detection of hepatocellular carcinoma (HCC) nodules.


A total of 100 treatment-naïve HCC patients admitted at our hospital between December 2007 and June 2009 were consecutively enrolled. The sensitivity and specificity of conventional and contrast-enhanced US were evaluated on a liver segment basis using dynamic CT as a reference standard. Movie files of conventional and enhanced US were stored separately for each segment (e.g., lateral, medial, anterior, and posterior) and reviewed randomly by two blinded readers.


A total of 138 HCC nodules (mean diameter 20.3 mm) were detected in 123 of 400 segments. Detection sensitivity of B-mode US was 0.837 for reader A and 0.846 for reader B, and that of CEUS was 0.732 for reader A and 0.831 for reader B. Specificity of B-mode US was 0.902 for reader A and 0.949 for reader B, and that of CEUS was 0.986 for reader A and 0.978 for reader B. CEUS false positives were mainly due to misidentification of hepatic cysts. A significant proportion of false-negative nodules are hyperechoic in B-mode US, likely because echogenicity hampers visualization of the defect in Kupffer imaging.


Kupffer imaging by CEUS with Sonazoid showed very high specificity but rather mediocre sensitivity for HCC detection. CEUS is highly suitable for confirmatory diagnosis of HCC; however, caution should be exercised in reaching a diagnosis based only on CEUS.


Hepatocellular carcinomaContrast-enhanced ultrasonographySonazoid





Contrast-enhanced ultrasonography

Conventional B-mode US

Conventional B-mode ultrasonography


Computed tomography


CT during arterial portography


CT during hepatic arteriography


Confidence interval


Gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid


Hepatitis B surface antigen


Hepatocellular carcinoma


Hepatitis C virus antibody


Mechanical index


The standards of reporting diagnostic accuracy


Superparamagnetic iron oxide in magnetic resonance imaging




Hepatocellular carcinoma (HCC) is a common malignancy worldwide, the incidence of which is increasing in the USA and elsewhere [14]. Prognosis of HCC is poor unless the cancer is detected and treated at an early stage. HCC develops usually in a liver already suffering from some chronic disease, most commonly cirrhosis. Thus, early diagnosis of HCC is essential in the management of patients with chronic liver disease [5]. While initial screening for HCC is usually performed by ultrasonography (US), contrast-enhanced dynamic computed tomography (CT) is used in a confirmatory role. Hyperattenuation in the arterial phase with washout in the late phase on dynamic CT is considered a definite sign of HCC [6]. However, dynamic CT is accompanied by substantial X-ray exposure, and examination may be contraindicated by allergy to contrast material or renal dysfunction.

Sonazoid (Daiichi-Sankyo, Tokyo, Japan), a recently introduced ultrasonographic contrast agent that contains perflubutane microbubbles within a shell of phosphatidylserine (diameter 2–3 μm), has been widely used in Japan since 2007 [7, 8]. When injected intravenously, Sonazoid particles reach the liver in about 15 s, allowing the hepatic arterial vascularity to be visualized (vascular imaging). In addition, Sonazoid is taken up by Kupffer cells approximately 15 min after intravenous injection [post-vascular phase (Kupffer imaging)], which enables differentiation of benign and malignant nodules [9]. In contrast to some other ultrasonographic contrast agents, perflubutane microbubbles continue to resonate with moderate ultrasound pressure without collapsing. Thus, Kupffer imaging is stable for more than several hours, facilitating whole-liver scanning [10]. Several reports have suggested that contrast-enhanced ultrasonography (CEUS) with Sonazoid could detect HCC nodules that are not detectable by conventional B-mode US [11, 12]. These features of Sonazoid raise the possibility of one-step depiction and diagnosis of HCC using ultrasonography alone or surveillance of HCC solely by CEUS. We sought in the present study to evaluate the diagnostic ability of CEUS in Kupffer imaging using Sonazoid for HCC.

Patients and methods

Study populations

The study design was based on the principles described in the standards of reporting diagnostic accuracy (STARD) initiative [13]. A total of 100 treatment-naïve HCC patients who were admitted to the authors’ institution for treatment between December 2007 and June 2009 were consecutively enrolled. Exclusion criteria were (1) patients with egg allergy or a past history of allergic reaction to Sonazoid; (2) those with HCC located in the caudate lobe; (3) those suffering severe cardiac or pulmonary dysfunction, which may affect Sonazoid delivery; and (4) those with HCC that showed an atypical dynamic CT pattern, as defined below. Data was collected prior to performance of Sonazoid CEUS.

This study was conducted according to the ethical guidelines for epidemiologic research of the Japanese Ministry of Education, Culture, Sports, Science and Technology and the Ministry of Health, Labour and Welfare. Informed consent was obtained from all patients. The study design was approved by the ethics committee of the authors’ institution.

Computed tomography

HCC was diagnosed by using dynamic CT prior to enrollment; hyperattenuation in the arterial phase with washout in the late phase was considered diagnostic of HCC (Fig. 1a, b) [14]. Dynamic CT was performed less than 1 month before Sonazoid CEUS. All CT examinations were performed using a 64-detector row helical CT (Aquilion 64; Toshiba Medical, Tokyo, Japan and LightSpeed VCT; GE Health Care, Tokyo, Japan). All subjects were given an intravenous injection of nonionic contrast material at 2.0 mL/kg body weight, to a maximum of 100 mL, using a power injector (Dual Shot GX; Nemotokyorindou, Tokyo, Japan) at a rate of 3 mL/s. We used fixed scanning delays in the double-arterial phase protocol: the early arterial phase starts at 25 s, the late arterial phase at 40 s, and the equilibrium phase starts 120 s after the beginning of contrast material injection. Following this, a bolus-tracking program is used to optimize the scanning delay for dynamic imaging.
Fig. 1

Typical appearance of HCC (arrow) on dynamic CT and unenhanced and enhanced US. Appearance of HCC on dynamic CT [early enhancement (a) with late washout (b)]. The nodule was detected as a hypoechoic lesion by unenhanced US (c) and visualized as an enhancement defect in Kupffer imaging Sonazoid CEUS (d)

Ultrasonographic imaging

Ultrasonography was performed before initiation of treatment on an inpatient basis after at least 5 h fasting using the SSA-770A (Aplio; Toshiba, Tokyo, Japan) ultrasound apparatus. Tissue harmonic imaging (2.5/5.0 MHz, 14–27 Hz) was used in B-mode US, and Sonazoid CEUS employed phase-inversion harmonic mode [mechanical index (MI), 0.2–0.3; gain, 75–90 dB; dynamic range, 45 dB; frame rate, 15 frames/s). The focus point was set 8–10 cm from the body surface (Fig. 1c).

Ultrasonographic contrast media

For Sonazoid CEUS, one vial (16 μg) of Sonazoid (Daiichi-Sankyo, Tokyo, Japan) was reconstituted with 2 mL of sterile water for injection, and 0.0075 mL/kg body weight of the solution was injected as a bolus via the antecubital vein and immediately flushed with 10 mL normal saline. Kupffer imaging was defined as the period starting 15 min after injection of Sonazoid (Fig. 1d).

Imaging technique and storage

US operators (EG, JI), each of whom had more than 8 years’ experience in liver ultrasonography, were blinded with regard to subjects’ clinical information including CT findings. All ultrasonographic images were recorded on movie files according to the following conditions. First, unenhanced ultrasonography was performed, and the images were stored in four separate movie files for each subject, one each for the left lateral, left medial, right anterior, and right posterior segments. Each segment was scanned from two directions in a standardized fashion, taking approximately 1 min. Sonazoid was then injected as described above, and after 15 min, Kupffer imaging scanning was performed and stored as for movie files. Vascular imaging images were not recorded. All movie files were formatted on uncompressed audio/video interleaving files with a resolution of 800 × 600 pixels and frame rate of 15 frames per second. Personal information was completely deleted from the movie files using the Toshiba Clip Washer ver. 3.5 software package (Toshiba Medical, Tokyo, Japan).

Movie file review

A total of 800 movie files, four each for unenhanced and Sonazoid-enhanced US per subject, were examined in a randomized order. Two readers (RT, RM), each of whom had more than 11 years’ experience in liver ultrasonography, reviewed each movie file independently. The readers were blinded to all clinical and demographic information. After reviewing each movie file, the reviewers recorded the number of HCC nodules in the file, if any, and the confidence of diagnosis was graded on a four-point scale: 1, definitely absent; 2, probably absent; 3, probably present; 4, definitely present. The location of detected nodules and frame timestamp were also recorded.

After all movie files had been evaluated, the file order was restored. The evaluation by each reader was compared with the findings of contrast-enhanced dynamic CT, using the latter as the reference standard. In some cases, nodules on a segment were also visualized in another movie file (e.g., a nodule in the medial segment was also visualized in a movie file for the anterior segment). In such cases, the third reader (EG) reassigned the score by referring to the readers’ comments (e.g., “definitely present; however, the nodule seems to be located in the medial segment rather than the anterior segment”). In cases where two or more nodules were located in the same segment, the third reader assessed whether the first and second readers accurately pointed out the nodule in the same segment for nodule-based subgroup analysis according to the readers’ comments.

To assess the characteristics of false-positive and -negative nodules, the third reader reviewed the corresponding movie files. Management of data regarding subject characteristics, scoring, and linking scores to randomized movie files was processed using Microsoft Access 2007 (Microsoft Corporation; Redmond, WA, USA).

Statistical analysis

Variables were expressed as mean ± standard deviations unless otherwise specified. Categorical variables were compared using Fisher’s exact probability test. Sensitivity and specificity of unenhanced and enhanced US in the detection of HCC nodules were calculated using the presence or absence of HCC in the corresponding liver segment on dynamic CT as the reference standard. Confidence intervals were calculated based on the F distribution. We also calculated positive and negative likelihood ratios defined as sensitivity/(1 − specificity) and (1 − sensitivity)/specificity, respectively. Differences in proportion were assessed using Fisher’s exact probability test. The concordance between the two readers was evaluated by weighted kappa statistics. The optimal cutoff in transforming the four-point scale to a dichotomous variable was validated by calculating the Youden index [15]. In addition to these segment-based analyses, tumor nodule-based analysis was also performed to evaluate tumor-associated factors that affected sensitivity for HCC detection. Statistical analyses were performed with S-plus ver. 7 (TIBCO Software, Inc., Palo Alto, CA, USA).


Subject characteristics and CT findings

The study population consisted of 60 men and 40 women, with a mean age of 67.5 years. Seventy-seven subjects were hepatitis C virus antibody (HCVAb)-positive, and nine were hepatitis B surface antigen (HBsAg)-positive. During the study period, no treatment-naïve HCC patients were excluded, and only one patient refused to participate in the study. The total number of HCC nodules detected was 138, with a mean diameter of 23.1 ± 7.7 mm determined by dynamic CT. Subject characteristics are summarized in Table 1 [16]. A total of 10 segments had two and three segments had three HCC nodules. Thus, 123 of 400 segments (13 lateral, 11 medial, 63 anterior, and 36 posterior segments) were HCC-positive, whereas the other 277 were negative.
Table 1

Baseline characteristics (n = 100)



Age (years)a

67.5 ± 10.6 (33–86)


 Male/female, n (%)/n (%)

60 (60.0)/40 (40.0)


 HBsAg positive, n (%)

9 (9.0)

 HCVAb positive, n (%)

77 (77.0)

Liver function

 Child-Pugh A/B, n (%)/n (%)

88 (88.0)/12 (12.0)

BMI (kg/m2)a

23.3 ± 3.1 (17.7–34.4)

Number of tumor nodules (n)a

1.38 ± 0.79 (1–4)

Maximum diameter of tumor (mm)a

23.1 ± 7.7 (8–41)

Serum albumin concentration (g/dL)a

3.79 ± 0.6 (2.4–4.7)

Total bilirubin concentration (mg/dL)a

0.97 ± 0.44 (0.4–2.4)

AFP level >100 ng/mL, n (%)

15 (15.0)

BMI body mass index

aData are expressed as mean ± SD (min–max)

Evaluation of movie files

Weighted kappa statistics related to the concordance of evaluations between the two readers were 0.732 for B-mode US and 0.718 for CEUS, indicating fair reproducibility. In subsequent analyses, scores of one or two were considered as HCC absent, and three or four as HCC present according to Youden index analysis.

HCC detection sensitivity

HCC detection sensitivity was evaluated in the 123 segments that contained HCC nodules (Table 2). Sensitivities of B-mode US were 0.837 [95% confidence interval (CI) 0.761–0.893] by reader A and 0.846 (95% CI 0.770–0.899) by reader B. Sensitivities of Sonazoid CEUS were 0.732 (95% CI 0.646–0.803) and 0.831 (95% CI 0.681–0.831), respectively. Reader A correctly detected HCC on unenhanced ultrasonography, but failed to detect the corresponding nodule on Sonazoid CEUS in 23 segments, and reader B did so in 18 segments.
Table 2

Sensitivity of HCC detection


Conventional US

Sonazoid CEUS

P value


95% CI


95% CI

Reader A









































Reader B









































Numbers in parentheses indicate the number of segments considered as positive by a reader/number of segments with HCC by dynamic CT

Factors affecting sensitivity were assessed based on HCC nodules (total of 138 nodules). Continuous variables were divided by the median value. When the two readers disagreed, a nodule was considered detected when the averaged four-point scale score exceeded two. As shown in Table 3, sensitivity was significantly lower for small nodules (<19 mm) with B-mode US, and a similar trend was observed with Sonazoid CEUS. Neither nodule depth from body surface nor subject BMI had a significant effect on the sensitivity of either B-mode US or Sonazoid CEUS. We then assessed the B-mode US echo pattern of a total of 101 nodules correctly detected by both readers on B-mode US; eight nodules were hyperechoic; the remaining 93 were hypoechoic. The readers failed to detect 15 of these on Sonazoid CEUS. Of these, four were hyperechoic and 11 were hypoechoic on Sonazoid CEUS, suggesting that detection of a defect in Kupffer imaging by Sonazoid CEUS was relatively difficult when the nodule was hyperechoic on B-mode US (P = 0.016 by Fisher’s exact test; Fig. 2).
Table 3

Factors affecting sensitivity


Conventional US

Sonazoid CEUS


95% CI

P value


95% CI

P value

Size of tumor (n)

 <19 mm (69)







 ≥19 mm (69)





Depth of tumor (n)

 <53 mm (69)







 ≥53 mm (69)





BMI (n)

 <23 kg/m2 (69)







 ≥23 kg/m2 (69)




Fig. 2

A false-negative detected by Sonazoid CEUS. A hypervascular nodule was detected in segment 8. The nodule was visualized as a hyperechoic lesion (arrow) by unenhanced US (a), but disappeared (circle) after Kupffer imaging by Sonazoid CEUS (b)

HCC detection specificity

A total of 277 segments did not contain HCC nodules, as determined by dynamic CT. The readers’ evaluation of movie files is summarized in Table 4. The specificity of B-mode US as evaluated by readers A and B was 0.902 (95% CI 0.861–0.932) and 0.949 (95% CI 0.917–0.970) and that of Sonazoid CEUS was 0.986 (95% CI 0.963–0.994) and 0.978 (95% CI 0.953–0.990), respectively. Sonazoid CEUS showed a higher specificity than B-mode US, irrespective of reader.
Table 4

Specificity of HCC detection


Conventional US




95% CI


95% CI

Reader A









































Reader B









































Numbers in parentheses correspond to the number of segments without HCC nodules by a reader/number of segments without HCC nodules by dynamic CT

Both readers made a false-positive evaluation with Sonazoid CEUS in three segments, and one reader did so in five other segments. Of these eight segments, five had a cystic lesion (as shown by dynamic CT), which may have been judged to be an enhancement defect in Kupffer imaging on Sonazoid CEUS (Fig. 3). Based on these sensitivities and specificities, the likelihood ratios for HCC presence when findings were positive were 11.4 and 43.4 with unenhanced and Sonazoid CEUS, respectively, and those for HCC presence when findings were negative were 0.171 and 0.222, respectively.
Fig. 3

A false-positive detected by Sonazoid CEUS. False-positive HCC detection by Sonazoid CEUS occurred mainly due to misrecognition of hepatic cysts (a, arrow) as HCC. Posterior enhancement was often obscured in Kupffer imaging by Sonazoid CEUS (b) and had an appearance (arrow) similar to that of an HCC nodule


The absence of Kupffer cells is one of the distinctive characteristics of hepatic malignant nodules, including HCC [8, 17]. This feature was first put to practical application for discriminating benign and malignant liver nodules using superparamagnetic iron oxide (SPIO) in magnetic resonance imaging (MRI) [18, 19]. Signal intensity of nodules on SPIO-MRI was reported to reflect grades of differentiation [20, 21]. The first-generation contrast agent Levovist (a suspension of galactose microparticle stabilized with palmitic acid) is also taken up by Kupffer cells in the liver and enables Kupffer cell imaging, similar to SPIO-MRI [22]. However, successful visualization of Kupffer cell defects by CEUS using Levovist requires sufficient ultrasound acoustic pressure to disrupt microbubbles. Repeated observation becomes impossible after microbubbles have been disrupted. This is a critical disadvantage in using Levovist CEUS for whole-liver screening. In contrast, Kupffer imaging using Sonazoid CEUS can be performed for several hours, which is suitable for imaging the entire liver.

In an assessment of the diagnostic accuracy of a test, evaluation of both sensitivity and specificity is essential. As an inverse correlation exists between sensitivity and specificity, reporting only sensitivity by recruiting those known to have the target disease is misleading and biased. Specificity of diagnostic imaging of liver nodules can be reported in at least three ways: (1) patient-, (2) segment-, or (3) nodule-based approaches. Calculation of specificity on a patient basis requires recruitment of individuals without HCC who undergo dynamic CT as a control. These controls should be patients with advanced liver fibrosis who are at a high risk for HCC. However, recruiting sufficient subjects to provide specificity with a narrow confidence interval is problematic. For a nodule-based study, controls should be benign nodules, such as hemangiomas, in patients with chronic liver disease. Previous reports on the diagnostic ability of Sonazoid CEUS on liver nodules were all based on this study design [11, 12, 2326]. However, the diagnostic accuracy assessed by this type of study is appropriate for differential diagnosis, but not for detection. In the present study we adopted the second approach. Ultrasound movie files were obtained from patients with known HCC, but readers did not know whether each segment contained a cancer nodule, which made evaluation of specificity possible. This design had the additional advantage that matching ultrasonographic conditions such as parenchymal heterogeneous echogenicity or subjects’ compliance was unnecessary, as “control” movie files were obtained from the same individuals. In addition, to reduce the likelihood of bias, ultrasound scanning of each segment was standardized, and B-mode US and Sonazoid CEUS movie files of the same segment were reviewed separately. These procedures were chosen for research purposes and obviously differ from clinical settings.

As a result, the sensitivity of Sonazoid CEUS for HCC detection was shown to be no greater than that of B-mode US. This may be due to several reasons. First, the lower acoustic power of Sonazoid CEUS compared with B-mode US (to avoid disrupting microbubbles) makes visualizing a deep nodule from the surface problematic. For enough sensitivity of CEUS, careful examination is demanded because CEUS is easily affected by the artifacts (such as bone and air) in low mechanical index (MI) mode. These artifacts were found especially in the right intercostal scan. Second, detection of nodules in obese patients is difficult due to attenuation of ultrasound by fatty tissue. However, subgroup analysis did not support this hypothesis. Hyperechoic nodules were unlikely to be detected by Kupffer imaging of Sonazoid CEUS, but this cannot entirely explain the low sensitivity. It should be noted that some nodules typical of HCC by dynamic CT are seen to possess Kupffer cells [27, 28]. However, Sonazoid CEUS can provide very stable post-vascular phase images for up to 60–120 min. Post-vascular phase obtained from later time (20–30 min) may increase the sensitivity of CEUS. The negative likelihood ratio of HCC presence by Sonazoid CEUS was insufficient, and thus caution should be taken when using this technique for diagnosis.

In contrast, the specificity of Sonazoid CEUS was shown to be about 98% irrespective of reader. The majority of false positives were due to misdiagnosis of cystic lesions as Kupffer imaging defects. Posterior enhancement, one of the major ultrasonographic characteristics of cystic lesions, may be obscured in Sonazoid CEUS. A positive likelihood ratio of at least 43 indicates that Sonazoid CEUS is suitable for confirmative diagnosis of HCC. Hyperechoic nodules were unlikely to be detected by post-vascular phase of Sonazoid US, but this cannot entirely explain the low sensitivity. Some new techniques can facilitate the diagnosis of tumor and should be combined with routine CEUS examination in clinical practice [29].

In clinical practice, US is performed in real time, so ambiguous lesions may be scrutinized by changing patients’ positions and requesting that they hold their breath. Final diagnoses are likely to be based on findings obtained by both contrasted and B-mode US. Thus, the sensitivity and specificity obtained in this study may be inferior to those in actual practice. Indeed, when movie files were re-examined 3 months after first review and images from both B-mode US and Sonazoid CEUS were examined simultaneously, the sensitivity improved compared to that for Sonazoid CEUS alone (data not shown). Other factors that may have affected HCC detectability were (1) that we did not use information in arterial images obtained by Sonazoid CEUS, which may have facilitated the assessment of the vascularity of targeted nodules; and (2) by using dynamic CT as the gold standard, atypical HCC detected with US may have been judged as false positives. The present study only included classical HCC and dynamic CT has its limitation. Early HCC is hypovascular on dynamic imaging in most cases. Its accurate diagnosis has remained difficult even with CT during hepatic arteriography (CTHA) and CT during arterial portography (CTAP). Gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)-enhanced MRI should be considered regarding diagnosis of early HCC [30].

In conclusion, we have shown that Sonazoid CEUS detects HCC with high specificity, but its sensitivity is no higher than that of B-mode US. Sonazoid CEUS is more suitable for confirmative diagnosis of HCC and can be performed immediately after B-mode US for this purpose.

Conflict of interest

The authors declare that they have no conflict of interest.

Copyright information

© Springer 2011