This study was reviewed and approved by the Ethics Review Board of Tokyo Medical University, and written informed consent was obtained from all participants. The diagnostic US scanner used in this study, the LOGIQ E10 (GE Healthcare, Wauwatosa, WI, USA), was supplied by the manufacturer. Authors with no conflicts of interest had full control over the inclusion of data and information.

Between June 2020 and January 2023, 219 patients with treatment-naïve hepatic lesions measuring 1 cm or larger were consecutively recruited in a prospective study at Tokyo Medical University Hospital. The inclusion criteria were as follows: (1) an age of at least 20 years, (2) at least one treatment-naïve hepatic lesion measuring 1 cm or larger, and (3) all lesions visible on baseline US. When multiple eligible lesions were detected in a patient, only one representative lesion was analyzed. Initially, 219 participants met the inclusion criteria. However, 38 participants were subsequently excluded for various reasons: 20 due to poor-quality images unsuitable for analysis, which were attributed to severe liver steatosis or lesions located deep within the liver; 10 due to deviation from the examination protocol; and eight due to incomplete diagnosis with CT/MRI. Thus, the study ultimately included 181 patients (108 male and 73 female), each with a single liver lesion. The median patient age was 68 years (interquartile range, 27 to 92 years), and the median size of the observed lesions was 25.0 mm (interquartile range, 17.0 to 38.5 mm). Of these lesions, 43.1% (78 of 181) were diagnosed as hepatocellular carcinomas (HCCs), 30.4% (55 of 181) as non-HCC malignancies, and 26.5% (48 of 181) as benign lesions. The non-HCC malignancies comprised 38 metastases, 15 intrahepatic cholangiocarcinomas, and two malignant lymphomas. The benign lesions included 25 hemangiomas, 18 focal nodular hyperplasias (FNHs), and five angiomyolipomas (AMLs).

Among the included lesions, 100% of non-HCC malignancies (55 of 181, 30.4%) were diagnosed histopathologically through surgery (n=5) or biopsy (n=50). Regarding HCCs, 76.9% (60 of 78) were diagnosed histopathologically (surgery, n=4; biopsy, n=56). The remaining 23.1% of HCCs (18/78) were non-invasively diagnosed, as they met the criteria for categorization as LR-5 according to the CT/MRI Liver Imaging Reporting and Data System version 2018 [9]. All FNHs and AMLs (23 of 181, 12.3%) were diagnosed histopathologically though biopsy (n=23). The authors’ institution routinely includes information on hepatic tumor pathology and immunohistochemistry in the pathological reports. All hemangiomas (25 of 181, 13.8%) exhibited typical MRI features, such as peripheral globular and centripetal enhancement or high signal intensity on T2-weighted images, which facilitated their identification as hemangiomas without pathological confirmation.

Conventional B-mode and CEUS examinations were performed by two hepatologists with 15 and 5 years of experience in abdominal US, respectively. The US scanner used was a LOGIQ E10 with a C1-6-D transducer (convex, 3.5-MHz center frequency). The CEUS imaging mode was configured to a fundamental/harmonic dual-display setting. For contrast harmonic imaging, the amplitude modulation method was employed with a low mechanical index, ranging from 0.16 to 0.2, and a dynamic range of 63 dB. The Sonazoid contrast agent was administered as a 0.5-mL bolus injection into the antecubital vein via a 21-gauge peripheral intravenous cannula, followed by a 10-mL saline flush. A timer was started at the moment of contrast agent injection. The targeted lesion was continuously recorded as a cine clip for 60 seconds after injection. During this time, patients were instructed to maintain gentle breathing. Subsequently, the same lesion was captured at 1-minute intervals as a 5-second cine clip during a breath-hold, from the 2-minute to the 10-minute mark post-injection. By examining the fundamental B-mode results, the target image with the maximum diameter was identified. The sequence of the CEUS protocol is depicted in Supplementary Fig. 1.

The WOPI technique utilizes a two-dimensional color map to illustrate the dynamics of washout timing within the CEUS protocol, as shown in Fig. 1. The color of each pixel on the map is derived from the intensity variation (in dB) observed during both the early portal venous phase and the vasculo-Kupffer phase of washout. These variations are mapped onto the x-axis (early portal venous phase) and y-axis (vasculo-Kupffer phase) of the color map. The resulting color scheme—with yellow indicating early portal venous washout, red signifying vasculo-Kupffer washout, and blue denoting minimal or no variation—provides a visual representation of washout timing across the lesion. For imaging analysis, three critical time frames were established: (1) the arterial phase, specifically the point of maximum lesion enhancement; (2) the early portal venous phase, occurring approximately 1-minute post-injection and marking the initial washout observation period; and (3) the vasculo-Kupffer phase, defined as occurring about 5 or 10 minutes after injection and enabling the visualization of any prolonged washout effects. This approach may facilitate a more nuanced analysis of washout characteristics, potentially improving the accuracy of FLL differentiation.

For the calculation of WOPI, the software first loads the previously recorded cine clips. The operator then selects three image frames, corresponding to the arterial phase, early portal venous phase, and vasculo-Kupffer phase of enhancement. For each image, the operator manually selects the cross-section of the target lesion at its largest diameter. For images captured between 0 to 1 minute after injection, the desired cross-section can be consistently obtained due to the regularity of the breathing cycle. For subsequent images, since patients are instructed to hold their breath, the desired cross-section is captured when the image is saved.

Initially, for the arterial phase peak enhancement frame, the operator defines the first region of interest (ROI) on the targeted liver lesion and the second ROI on the adjacent liver parenchyma. ROIs of the same size are then automatically applied to subsequent frames in the same location. The operator must confirm that the ROI is properly positioned on the target or adjust its location to the correct position if displacement occurs due to motion.

Following ROI selection, the software calculates the WOPI by quantifying the differences in pixel intensity between the lesion and the average intensity within the parenchymal tissue ROI. This process mirrors the conventional analytical methods employed in CEUS assessments. The software then superimposes the calculated differences (expressed in dB) for the early portal venous and vasculo-Kupffer phases onto the contrast B-mode image, producing a visual representation of the washout characteristics. This facilitates not only the detection of washout presence or absence but also a semi-quantitative assessment of the washout degree through the brightness and hue of the color. The determination of ROI settings is made collaboratively by two physicians via consensus, thus promoting the reliability and accuracy of the analysis.

A receiver operating characteristic (ROC) observer study was independently conducted by four hepatologists (H.T, Y.Y, H.T, T.W, with 12, 10, 3, and 1 year of experience in liver CEUS, respectively) who were not involved in performing the CEUS examinations and were blinded to the final diagnoses and patient information. They used ROC software (ROCViewer-ForMethod 1 ver. 1.0.1) developed by the Japanese Society of Radiological Technology task group [10]. This software featured a rating bar that allowed the observers to indicate their confidence in differentiating between benign and malignant lesions by clicking on the bar, with the left and right ends representing "definitely benign" and "definitely malignant," respectively (Supplementary Fig. 2). The ratings were quantified on a continuous scale from 0.0 to 1.0, based on the distance from the left end of the bar to the selected point.

Four independent reading sessions were conducted, featuring the following image sets: (1) native CEUS images comprising B-mode, arterial phase (peak enhancement), 1-minute, and 5-minute images; (2) a B-mode image paired with a washout-parametric image derived from the peak, 1-minute, and 5-minute images; (3) native CEUS images including B-mode, arterial phase (peak enhancement), 1-minute, and 10-minute (Kupffer phase) images; and (4) a B-mode image alongside a washout-parametric image generated from the peak, 1-minute, and 10-minute images. Initially, the four readers independently evaluated sets 1 and 2 in sequence. Subsequently, after a 1-month interval to mitigate any potential learning effects, the same readers independently assessed sets 3 and 4 in sequence.

For the observer study, 90 patients were randomly selected. These included 30 with HCCs, 30 with non-HCC malignancies (comprising 25 liver metastases and five intrahepatic cholangiocarcinomas), and 30 with benign lesions (18 hemangiomas and 12 FNHs). This group was selected from the imaging database, which included all 181 patients and 181 lesions. The decision to limit the number of cases for the observer study was based on two main considerations. First, a complete review of all lesions would have excessively burdened the observers, and reducing the number of cases helped maintain their concentration. Second, the lesion database was imbalanced, with 133 of 181 lesions (73.5%) malignant and only 48 benign, making it unsuitable to conduct the observer study with the entire set. Furthermore, malignant lymphomas and AMLs were excluded from the observer study due to their rarity. The details are presented in Table 1. Prior to the observer study, each reader was trained on how to use the software’s rating bar and interpret the WOPI, with the aid of some pretest cases. The diagnostic criteria for CEUS used by the readers to determine their confidence in differentiating benign from malignant lesions were based on established US criteria for hepatic tumors [11]. The WOPI was interpreted using specific reading rules: lesions were classified as malignant if they displayed red or yellow hues on the color map (indicative of a washout pattern) and as benign if they exhibited blue hues. One hepatologist (K.S.) who did not participate in the observer study reviewed all FLLs and documented the intensity and pattern of arterial phase enhancement, as well as the presence or absence of washout, up to 10 minutes post-injection.

To summarize the data, normally distributed continuous variables are expressed as means±standard deviations, while non-normally distributed continuous variables are presented as medians (interquartile ranges). Categorical variables are expressed as absolute numbers with percentages. The differences in background characteristics between malignant and benign lesions were assessed using the Student t-test or the chi-square test. The overall diagnostic performance of CEUS and WOPI for differentiating between benign and malignant FLLs was evaluated using the area under the receiver operating characteristic curve (AUC) and the 95% confidence interval (CI). Cutoff values for the two techniques were determined to maximize the Youden index, as sensitivity and specificity were equally important. Sensitivity, specificity, and accuracy, along with their 95% CIs, were reported for these cutoff values. The average AUC values were compared using the DeLong test, and pooled sensitivity, specificity, and accuracy values were compared using the McNemar test. The intraclass correlation coefficient (ICC) was used to evaluate interobserver agreement for both techniques, with ICC values greater than 0.75 considered indicative of good agreement [12]. All statistical analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) and JSRT-MRMC (multi-reader multi-case) software [13]. P-values of less than 0.05 were considered to indicate statistical significance.

The diagnostic performance of cCEUS and WOPI was evaluated based on the average AUC and pooled sensitivity, specificity, and accuracy metrics, as observed by four readers (Table 2). The data for each of the four readers are provided in Supplementary Table 1. For the 5-minute evaluation, cCEUS exhibited an average AUC of 0.87 and pooled sensitivity of 0.83 (95% CI, 0.78 to 0.88), specificity of 0.83 (95% CI, 0.75 to 0.89), and accuracy of 0.83 (95% CI, 0.79 to 0.87). WOPI showed similar performance, with an average AUC of 0.85 and pooled sensitivity of 0.80 (95% CI, 0.75 to 0.85), specificity of 0.78 (95% CI, 0.70 to 0.85), and accuracy of 0.80 (95% CI, 0.75 to 0.84). For the 10-minute evaluation, cCEUS metrics were slightly lower, with an average AUC of 0.84 and pooled sensitivity of 0.78 (95% CI, 0.73 to 0.83), specificity of 0.82 (95% CI, 0.74 to 0.88), and accuracy of 0.79 (95% CI, 0.75 to 0.84). The performance of WOPI was comparable to that of cCEUS, with an average AUC of 0.83 and pooled sensitivity of 0.82 (95% CI, 0.76 to 0.86), specificity of 0.77 (95% CI, 0.68 to 0.84), and accuracy of 0.80 (95% CI, 0.76 to 0.84).

Statistical analysis revealed no significant difference in diagnostic performance between cCEUS and WOPI at either 5 or 10 minutes. The differences in the AUC were 0.0183 (95% CI, -0.0533 to 0.0898) at 5 minutes and 0.0062 (95% CI, -0.0161 to 0.0285) at 10 minutes. Furthermore, extending the observation period from 5 to 10 minutes did not significantly affect the diagnostic outcomes for either technique, with the difference in AUC with being 0.0238 (95% CI, -0.0240 to 0.0716) for cCEUS and 0.0117 (95% CI, -0.0623 to 0.0858) for WOPI.

The interobserver agreement for both cCEUS and WOPI was analyzed, with results reported for the initial 5-minute and extended 10-minute evaluation periods. For the 5-minute evaluation, cCEUS images exhibited an ICC of 0.71 (95% CI, 0.64 to 0.79), indicating moderate agreement among readers. WOPI images showed an ICC of 0.77 (95% CI, 0.70 to 0.83), reflecting good agreement (Table 3). For the 10-minute evaluation, cCEUS images maintained moderate agreement with an ICC of 0.67 (95% CI, 0.58 to 0.75). WOPI images again demonstrated good agreement with an ICC of 0.77 (95% CI, 0.70 to 0.90) (Table 3). This analysis suggests that WOPI yielded higher inter-reader agreement than cCEUS at both the 5-minute and 10-minute marks, indicating that WOPI may offer a more reliable basis for the differential diagnosis of FLLs.

The arterial phase and washout characteristics are detailed in Tables 4 and 5, respectively. Of the 18 hemangiomas studied, 16 lesions (88.9%) exhibited the typical peripheral globular enhancement pattern with centripetal filling. Conversely, two lesions (11.1%) displayed diffuse hyperenhancement during the arterial phase. Regarding washout characteristics, six lesions (33.3%) exhibited hypoenhancement and 12 lesions (66.7%) displayed isoenhancement in the Kupffer phase. Notably, one lesion that presented with diffuse hyperenhancement in the arterial phase also showed early washout, occurring within 60 seconds.

Among the 12 FNHs, all nodules showed diffuse and homogeneous enhancement in the arterial phase. Among them, one lesion (8.3%) exhibited early washout (<60 seconds). In the Kupffer phase, eight lesions (66.7%) showed isoenhancement, three (25.0%) showed hyperenhancement, and one (8.3%) showed hypoenhancement.

Among the 30 HCCs, 18 lesions (60.0%) exhibited typical arterial phase hyperenhancement followed by mild washout during the vasculo-Kupffer phase (including the Kupffer phase). Four lesions (13.3%) demonstrated arterial phase hyperenhancement with persistent isoenhancement up to the Kupffer phase. Three lesions (10.0%) displayed arterial phase hyperenhancement with early washout (<60 s). Four lesions (13.3%) showed isoenhancement in the arterial phase; of these, two showed mild washout in the vasculo-Kupffer phase (including the Kupffer phase), while the other two exhibited early washout. One lesion (3.3%) presented with rim enhancement during the arterial phase and washout during the vasculo-Kupffer phase, including the Kupffer phase.

Among the 30 non-HCC malignancies (25 liver metastases and five intrahepatic cholangiocarcinomas), a typical pattern of rim enhancement was observed in nine lesions (30.0%), while diffuse enhancement was observed in 21 (70.0%). Early washout (<60 seconds) was seen in 27 lesions (90.0%). The remaining three lesions (10.0%), all of which were metastases, exhibited washout by 5 minutes.