Kupffer-phase Sonazoid ultrasound LI-RADS for liver cancer: diagnostic performance and algorithm modification

Article information

Ultrasonography. 2026;45(2):184-193

Publication date (electronic) : 2026 February 23

doi : https://doi.org/10.14366/usg.25148

Xuejun Chen^1,2,*, Daohui Yang^1,*, Lewu Lin^1,*, Aiqin Wu^1,*, Linjin Huang¹, Lijuan Mao³, Hansheng Xia³, Qing Lu^,1,3,4

¹Department of Ultrasound, Zhongshan Hospital (Xiamen), Fudan University, Xiamen, China

²Clinical Research Center for Precision Medicine of Abdominal Tumor of Fujian Province, Xiamen, China

³Department of Ultrasound, Zhongshan Hospital, Fudan University, Shanghai, China

⁴Xiamen Clinical Research Center for Cancer Therapy, Xiamen, China

Correspondence to: Qing Lu, PhD, Department of Ultrasound, Zhongshan Hospital, Fudan University, Shanghai 200030, China Tel. +86-21-64041990 Fax. +86-21-64038472 E-mail: lu.qing@zs-hospital.sh.cn

*These authors contributed equally to this work.

Received 2025 August 6; Revised 2026 January 6; Accepted 2026 January 12.

Abstract

Graphical abstract

Introduction

Hepatocellular carcinoma (HCC), which accounts for more than 80% of primary liver malignancies, remains a major global health burden. Unique among malignancies, HCC can be diagnosed without invasive procedures based on typical imaging characteristics in patients at high risk [1]. Although international guidelines from the American Association for the Study of Liver Diseases and the European Association for the Study of the Liver prioritize contrast-enhanced computed tomography (CT) or magnetic resonance imaging (MRI), contrast-enhanced ultrasound (CEUS) remains a valuable supplementary diagnostic tool in most guidelines [2–4]. To standardize CEUS interpretation, the American College of Radiology (ACR) released the CEUS Liver Imaging Reporting and Data System (LI-RADS), which integrates vascular-phase imaging findings—such as non-rim arterial phase hyperenhancement (APHE) and late-onset washout (≥60 seconds)—to categorize liver nodules in patients at high risk [5]. However, CEUS LI-RADS v2017 does not fully leverage the diagnostic potential of Kupffer cell–specific agents such as Sonazoid [6].

Sonazoid, a second-generation ultrasound contrast agent, enables visualization of both the vascular and Kupffer phases. Because most malignant hepatic lesions lack Kupffer cells, Kupffer-phase (KP) hypoenhancement has emerged as a valuable indicator of malignancy [7]. Accordingly, recent guidelines from the Japan Society of Hepatology (JSH) and the Korean Liver Cancer Association–National Cancer Center (KLCA-NCC) have incorporated KP hypoenhancement as a major diagnostic criterion for HCC [8,9]. However, the optimal integration of Sonazoid into the existing CEUS LI-RADS framework remains unclear. Key questions include the diagnostic weight of KP hypoenhancement, the optimal timing for washout assessment given Sonazoid’s prolonged vascular retention, and how modified LI-RADS approaches compare with established guidelines that already incorporate KP findings [10–14].

This study aimed to refine the Sonazoid-based CEUS LI-RADS algorithm by incorporating KP features and revising the washout time window. The diagnostic performance of the modified algorithms was also compared with the JSH and KLCA-NCC guidelines to evaluate clinical utility in regions where locoregional therapy is prioritized.

Materials and Methods

Study Participants

From August 2022 to September 2024, a total of 1,484 patients underwent Sonazoid-enhanced CEUS for liver nodules at Zhongshan Hospital (Shanghai) and its Xiamen Branch.

The inclusion criteria were as follows: (1) solitary liver nodule with histopathologic confirmation; (2) treatment-naive status (no prior local or systemic therapy); (3) lesion size ≥10 mm on grayscale ultrasound; (4) high risk for HCC according to LI-RADS, including cirrhosis of any cause (e.g., chronic hepatitis B virus [HBV] or hepatitis C virus infection), chronic HBV infection without cirrhosis, or prior HCC; and (5) Sonazoid-enhanced CEUS performed within 2 weeks before surgery or biopsy.

The exclusion criteria were as follows: (1) poor CEUS image quality (e.g., severe fatty liver, deep nodule location) or incomplete video recording; and (2) evidence of macrovascular or bile duct invasion on preoperative cross-sectional imaging (CT, MRI, or ultrasound). A detailed patient selection flowchart is provided in Fig. 1.

Fig. 1.

Flowchart of patient recruitment.

CEUS, contrast-enhanced ultrasound; HCC, hepatocellular carcinoma.

Image Interpretation and Categorization

CEUS features were assessed across dynamic phases, including the arterial phase (0–30 seconds), and at predefined time points after injection (1, 2, 5, and 10 minutes) to evaluate progressive washout and KP hypoenhancement [12–15]. Algorithms were applied using post-injection washout assessment windows of 2, 5, and 10 minutes. Two experienced radiologists (D.Y. and Q.L. each with more than 10 years of liver CEUS experience) independently evaluated all lesions using the ACR CEUS LI-RADS v2017 criteria [5], including: (1) peak enhancement (hypo-, iso-, or hyperenhancement); (2) arterial phase pattern (rim, non-rim, or peripheral globular); (3) washout timing (early [<60 seconds] or late [≥60 seconds]); and (4) washout degree (mild [enhancement less than that of the liver but not devoid of enhancement] or marked [black or punched-out appearance]). Discrepancies were adjudicated by a third senior reviewer (H.X., with more than 20 years of liver CEUS experience).

A fourth blinded reviewer (X.C., more than 5 years of liver CEUS experience) independently assigned LI-RADS categories using three predefined CEUS LI-RADS algorithms based on washout assessments at 2, 5, and 10 minutes after injection (Fig. 2). Additionally, two KP-integrated algorithms—Modified LI-RADS-5min and modified LI-RADS-10min—were constructed by incorporating KP hypoenhancement into the corresponding vascular-phase–only frameworks (CEUS LI-RADS-5min and CEUS LI-RADS-10min, respectively). Diagnostic performance was also compared with the JSH 2021 and KLCA-NCC 2023 guidelines. Further details of all diagnostic algorithms are provided in Supplementary Data 1. All reviewers were blinded to the patients’ medical history, other imaging results, and histopathological data.

Fig. 2.

A 58-year-old man with a >20-year history of chronic hepatitis B virus infection.

A. Grayscale ultrasound shows a hypoechoic liver nodule measuring 48 mm in maximum diameter (arrows), later confirmed as hepatocellular carcinoma on histopathology. B. Contrast-enhanced ultrasound shows non-rim arterial phase hyperenhancement (arrows). C. No washout is observed at 2 minutes after contrast injection (arrows). D, E. The nodule shows mild washout at 5 minutes and marked washout at 10 minutes after injection, respectively (arrows).

Results

Baseline Characteristics

In total, 558 patients (mean age, 62.2±11.4 years; 409 men and 149 women) with solitary liver nodules were included. The cohort comprised 482 HCCs, 66 non-HCC malignancies (including 28 intrahepatic cholangiocarcinomas [ICCs], 13 combined hepatocellular-cholangiocarcinomas, and 25 metastatic liver tumors), and 10 benign lesions (two dysplastic nodules, three focal nodular hyperplasias [FNHs], two angiomyolipomas, two hemangiomas, and one inflammatory pseudotumor). All diagnoses were pathologically confirmed by surgical resection (n=427) or percutaneous biopsy (n=131). Baseline characteristics of the participants are summarized in Table 1.

Table 1.

Baseline characteristics of participants and liver nodules

Characteristics of CEUS Features

Table 2 summarizes the CEUS features of the liver nodules. Non-rim APHE was identified in 91.1% of HCCs, 51.5% of non-HCC malignancies, and 80.0% of benign lesions. Mild washout in HCCs was detected at 2 minutes (61.4%, 296/482), 5 minutes (89.2%, 430/482), and 10 minutes (83.6%, 403/482). The prevalence of marked washout in HCCs increased over time (0.6% at 2 minutes, 3.5% at 5 minutes, and 13.3% at 10 minutes). For non-HCC malignancies, the detection rates of mild washout were 66.7% (44/66), 72.7% (48/66), and 45.5% (30/66) at 2, 5, and 10 minutes, respectively, whereas the rates of marked washout were 12.1% (8/66), 18.2% (12/66), and 45.5% (30/66) at the same time points. No benign lesions exhibited marked washout. However, 80.0% (8/10) demonstrated mild washout at one or more time points, whereas 20.0% (2/10) showed no washout throughout the CEUS examination.

Table 2.

Individual CEUS features of liver nodules on Sonazoid-enhanced CEUS

Diagnostic Performance of CEUS LI-RADS

Detailed LI-RADS category assignments based on different washout time windows are provided in Supplementary Tables 2–5. As shown in Table 3 and Fig. 3, CEUS LI-RADS-5min and CEUS LI-RADS-10min demonstrated comparable diagnostic performance (AUC, 0.745 vs. 0.734; P=0.095), and both outperformed CEUS LI-RADS-2min (AUC, 0.624). Specificity and positive predictive value (PPV) were high for the CEUS LI-RADS-5min and CEUS LI-RADS-10min algorithms (specificity, 84.2% for both; PPV, 96.3% and 97.4%, respectively), whereas sensitivity remained modest (64.7% and 62.7%, respectively).

Table 3.

Comparison of HCC diagnostic performance of CEUS LI-RADS using different washout time windows

Fig. 3.

Receiver operating characteristic curves for hepatocellular carcinoma (HCC) diagnosis using contrast-enhanced ultrasound (CEUS) Liver Imaging Reporting and Data System (LI-RADS) with different washout time.

Receiver operating characteristic curves are shown for HCC diagnosis using the CEUS LI-RADS with different washout time windows (A) and modified CEUS LI-RADS compared with the Japan Society of Hepatology (JSH) and Korean Liver Cancer Association (KLCA) guidelines (B). The modified CEUS LI-RADS-10min algorithm presents the highest area under the curve (AUC; 0.782), outperforming the other criteria.

Diagnostic Performance of Modified CEUS LI-RADS

The modified CEUS LI-RADS-10min algorithm incorporating KP hypoenhancement significantly improved diagnostic performance compared with CEUS LI-RADS-10min (AUC, 0.782 vs. 0.734; P=0.016) (Table 4, Fig. 3). In contrast, the modified CEUS LI-RADS-5min algorithm showed no significant difference (AUC, 0.695 vs. 0.745; P=0.075). The modified CEUS LI-RADS-10min algorithm correctly reclassified 98 additional HCCs, including one LR-4 nodule (APHE without washout) (Fig. 4) and 97 LR-M nodules (APHE with early washout) (Fig. 5). However, it misclassified eight non-HCC lesions (two FNHs and six ICCs) as LR-5. The six ICCs were upgraded from LR-M to LR-5 because of overlapping CEUS features, including APHE, early washout, and mild KP hypoenhancement. The two misclassified FNHs showed APHE with mild KP hypoenhancement, resulting in reclassification from LR-4 to LR-5 under the modified CEUS LI-RADS-10min algorithm.

Table 4.

Diagnostic performance for HCC using different CEUS criteria

Fig. 4.

A 45-year-old man with chronic hepatitis B virus infection.

A. Grayscale ultrasound shows a 27-mm hypoechoic nodule in the right hepatic lobe. B, C. Contrast-enhanced ultrasound (CEUS) shows non-rim arterial phase hyperenhancement (APHE) at 20 seconds after injection (arrows), with no definite washout observed up to 5 minutes. According to CEUS Liver Imaging Reporting and Data System (LI-RADS) v2017, the lesion was classified as LR-4. D. At 10 minutes after injection (Kupffer phase), the nodule shows mild washout. Under the modified CEUS LI-RADS-10min algorithm, this “APHE with no washout” lesion was correctly upgraded from LR-4 to LR-5. The diagnosis of hepatocellular carcinoma was subsequently confirmed by histopathology.

Fig. 5.

A 63-year-old woman with hepatitis B virus–related cirrhosis.

A. Grayscale ultrasound shows a 43-mm isoechoic to hyperechoic nodule in the right hepatic lobe, initially categorized as LR-M according to contrast-enhanced ultrasound (CEUS) Liver Imaging Reporting and Data System (LI-RADS) v2017. B. The nodule demonstrates non-rim arterial phase hyperenhancement (APHE) at 22 seconds after injection (arrows). C. Early mild washout is observed at 52 seconds post-injection (arrows), consistent with the LR-M category per CEUS LI-RADS v2017. D. Mild washout persists during the late phase at 5 minutes post-injection (arrows). E. Mild washout persists during the Kupffer phase at 10 minutes post-injection (arrows). Under the modified CEUS LI-RADS-10min criteria, this “APHE with early washout” nodule was correctly reclassified from LR-M to LR-5. The final diagnosis of hepatocellular carcinoma was confirmed by histopathology.

Discussion

This two-center study evaluated a modified CEUS LI-RADS algorithm that incorporates KP imaging features and optimizes washout timing for diagnosing HCC using Sonazoid-based CEUS. The modified CEUS LI-RADS-10min algorithm improved diagnostic sensitivity while maintaining reasonable specificity and PPV, providing a more balanced diagnostic approach. These findings support the potential of a KP-integrated LI-RADS framework to refine Sonazoid CEUS interpretation and promote standardization.

Washout timing plays a critical role in CEUS LI-RADS categorization; however, the optimal time window for washout assessment with Sonazoid remains a topic of debate. This study showed that CEUS LI-RADS-5min and CEUS LI-RADS-10min had comparable diagnostic performance (AUC, 0.745 vs. 0.734; P=0.095), and both outperformed CEUS LI-RADS-2min (AUC, 0.624). These results challenge the previous assumption that delayed imaging inherently improves the diagnostic performance of Sonazoid-enhanced CEUS [11,12,14–17]. Prior studies have reported conflicting findings: Takahashi et al. [11] and Hwang et al. [12] observed better sensitivity at 10 minutes, whereas Kang et al. [18,19] reported that the 5-minute criterion achieved slightly higher diagnostic performance than the 10-minute criterion (72.4%; 95% confidence interval, 64.1% to 79.3%) and that extending the observation period beyond 6 minutes provided no incremental diagnostic benefit. The present results are consistent with Kang’s findings, indicating that classification criteria, rather than washout timing alone, play a pivotal role in diagnostic performance.

Despite high specificity and PPV, CEUS LI-RADS-5min and CEUS LI-RADS-10min displayed suboptimal sensitivity, misclassifying more than one-third of HCCs as non-LR-5. To address this limitation and better align with clinical needs, CEUS LI-RADS-10min was modified by reclassifying LR-4 nodules with KP hypoenhancement as LR-5 and LR-M nodules with early washout and mild KP hypoenhancement as LR-5. As a result, one LR-4 nodule and 97 LR-M nodules were correctly reclassified as LR-5, yielding the highest diagnostic performance for HCC (AUC, 0.782) through improved sensitivity (82.8%) while maintaining reasonable specificity (73.7%), comparable to the results of Li et al. [13]. In contrast, the modified CEUS LI-RADS-5min algorithm achieved a significantly lower AUC than the modified CEUS LI-RADS-10min algorithm (AUC, 0.695 vs. 0.782; P<0.001), with a specificity of only 55.3%.

Recent efforts have focused on modifying CEUS LI-RADS using various approaches, including revising the early washout threshold [20,21]. Huang et al. [20] reported that shortening the early washout cutoff from 60 to 45 seconds increased the sensitivity for diagnosing HCC (52.1% vs. 68.8%, P=0.004) without a significant reduction in specificity (88.0% vs. 76.0%, P=0.118). In the present study, six ICCs were misclassified as LR-5 because they showed early washout (40–50 seconds) together with mild KP hypoenhancement. These nodules might be correctly categorized under a revised early washout threshold, consistent with Huang’s findings. However, interobserver agreement for early washout remained only moderate (κ=0.580), underscoring its subjectivity. Therefore, modifications based on this feature warrant further investigation, potentially incorporating quantitative time-intensity curve analysis to improve interobserver agreement.

Additionally, the two misclassified FNHs displayed APHE and no washout during the vascular phase, while showing mild KP hypoenhancement. Certain benign lesions, such as FNH with central scarring or hemangiomas with fibrotic components, have been reported to exhibit enhancement patterns that mimic mild washout on CEUS [22,23]. However, the two FNHs in this cohort demonstrated overall KP hypoenhancement rather than stellate central hypoenhancement; therefore, these findings were not considered consistent with a central scar. Both lesions were located in moderately steatotic livers, in which background steatosis likely increased lesion-to-parenchyma contrast, creating the visual impression of mild washout. A fatty liver background may influence interpretation of washout characteristics on Sonazoid-enhanced CEUS [24]. Because all cases were pathologically confirmed, the study design inevitably included some lesions with atypical enhancement patterns. Accordingly, these observations may not fully represent the spectrum of imaging appearances of FNH.

The clinical utility of CEUS has been recognized in major guidelines, including those from the JSH and KLCA, both of which incorporate Sonazoid as a complementary tool for diagnosing HCC. In the present study, the modified CEUS LI-RADS-10min algorithm showed diagnostic performance comparable to that of the KLCA-NCC guideline and significantly higher than that of the JSH guideline. Although the JSH guideline achieved the highest sensitivity (91.7%), its specificity was low (53.9%) because any nodule demonstrating APHE and washout—regardless of timing or degree—is classified as HCC. In contrast, the modified CEUS LI-RADS-10min algorithm categorized nodules with early or marked washout as LR-M, substantially improving specificity (73.7%) while maintaining high sensitivity (82.8%). The KLCA guideline, which excludes nodules with early washout from an HCC diagnosis, had significantly lower sensitivity (68.3%) in this cohort because 24.7% of HCCs exhibited early washout. By incorporating mild KP hypoenhancement, the modified CEUS LI-RADS-10min algorithm reclassified most of these early-washout lesions as LR-5, thereby achieving a more balanced diagnostic profile while aligning well with the KLCA guideline. Notably, the modified CEUS LI-RADS-10min algorithm is intended as a complementary approach rather than a replacement for existing guidelines. In clinical practice, CEUS is commonly used to further characterize indeterminate lesions on CT or MRI, where specificity is prioritized. This balanced approach may improve lesion detection in high-risk patients without compromising specificity, supporting its incorporation into future CEUS LI-RADS frameworks and improving standardization of Sonazoid-based CEUS.

The present study has several limitations. First, the use of pathology as the sole reference standard may have introduced verification bias, as resected or biopsied nodules are more likely to be malignant and assigned higher LI-RADS categories. In addition, the retrospective design and the high prevalence of HCC (86.4%)—reflecting a population largely selected for surgical resection or biopsy confirmation—could have overestimated diagnostic performance compared with screening or surveillance cohorts. Consequently, the high pretest probability of malignancy could inflate the PPV while reducing the negative predictive value, limiting the generalizability of these findings to broader clinical settings. Second, although interobserver agreement for KP hypoenhancement and washout assessment was substantial, these evaluations remain partly subjective. Third, nonviral HCC etiologies were underrepresented. Prospective, multicenter validation in more diverse populations, along with further standardization of KP interpretation, is warranted.

In conclusion, incorporating KP imaging features and optimizing washout timing improved the performance of CEUS LI-RADS for diagnosing HCC. The modified CEUS LI-RADS-10min algorithm achieved higher sensitivity while retaining reasonable specificity compared with CEUS LI-RADS based on a vascular-phase contrast agent, offering a balanced and practical framework for interpreting Sonazoid-based CEUS. This approach provides a clinically feasible balance between sensitivity and specificity and may inform future refinements of CEUS-based classification systems, facilitating standardization of Sonazoid-enhanced CEUS in clinical practice.

Supplementary Material

Article information Continued

Notes