Between August 2018 and December 2018, patients with HCC treated at a tertiary referral center were consecutively enrolled in this study. Patients were included in the study if they were ≥20 years old, had underlying liver cirrhosis, and had definite HCC on CT/MRI or viable HCC on follow-up CT/MRI performed after treatment for HCC [5]. Patients were excluded if they refused to participate or if the ultrasound evaluation of the tumor was limited because of poor tumor visibility due to deep location or a marked motion artifact (Fig. 1).

A radiologist (J.S.B., with 7 years’ experience in abdominal ultrasound examinations) performed all CDI, PDI, and MFI examinations using an ultrasound scanner (EPIQ 7, Philips Healthcare, Eindhoven, Netherlands) with a 1-5 MHz convex probe. Initially, the known HCC was located using B-mode ultrasound and referring to the previously acquired contrast-enhanced CT or MRI images. After locating the HCC, the best sonic window was obtained, and the tumor size and depth were measured. Subsequently, the patient was asked to hold his/her breath, and the color Doppler mode was turned on. Color gain, scale, and wall filter were adjusted for each tumor to acquire the best images. The sampling window was placed to include the tumor and surrounding liver parenchyma. A cine loop of approximately 15-20 seconds was stored while slowly sweeping the HCC in the craniocaudal direction without applying pressure. Additionally, at least five still images of the HCC were captured where the vascularity on CDI was best demonstrated. Next, the ultrasound mode was sequentially converted to power Doppler mode, and then to MFI mode. A cine loop and at least five still images were obtained from each mode. The same sonic window was used when performing all the above ultrasound examinations.

Two radiologists (both with 6 years’ experience in abdominal ultrasound examinations) who did not perform the ultrasound examinations independently reviewed the CDI, PDI, and MFI image sets on a picture archiving and communication system (Infinitt PACS, Infinitt Healthcare, Seoul, Korea). The reviewers were informed that the images had been obtained from patients with HCC but were blinded to other clinical information. The order of reviewing the three image sets was at the reviewers’ discretion; however, the review of each image set (CDI, PDI, or MFI) was performed with at least a 2-week interval to reduce recall bias. The reviewers assessed both cine loops and still images and graded the presence and pattern of vascularity. The presence of vascularity was graded on a 5-point scale: 0, absent; 1, a few spotty signals in the tumor; 2, signals in <25% of the tumor; 3, signals in 25%-50% of the tumor; and 4, signals in >50% of the tumor [16]. A score of 3 or higher was considered as positive for the presence of vascularity. The pattern of vascularity was chosen from the following categories: basket pattern, vessels in tumor, spot pattern, detouring pattern, a combination of the above, or others [17]. After an independent review, a consensus between the two reviewers was reached. Any discrepancies were resolved by discussion with a senior radiologist (with 27 years’ experience in abdominal ultrasound examinations).

Arterial-phase images of CEUS examinations were used as the reference standard for the vascularity of the HCCs because CEUS has the highest spatial and temporal resolution for the depiction of tumoral vascularity. One of the two radiologists (J.M.L. or J.S.B.) performed the CEUS examination immediately after completing the precontrast ultrasound examinations. For the CEUS examination, a single 2.4-mL bolus of prepared sulfur hexafluoride-filled microbubbles (SonoVue, Bracco, Milan, Italy) was manually injected into the antecubital vein, followed by a 10-mL 0.9% saline flush. The following parameters were used during the CEUS examinations: mechanical index, 0.08; dynamic range, 65 dB; 10-13 frames per second; signal persistence turned on; power modulated pulse inversion technology; echo-signal gain below noise visibility; and a single focus at the HCC. After injection of the contrast media, a cine loop of the target tumor was recorded for the first 60 seconds, which was followed by intermittent scans to minimize the disruption of microbubbles for up to 5 minutes. The CEUS imaging plane was carefully selected to be the same as those used for the CDI, PDI, and MFI examinations. After performing the CEUS examinations, the two radiologists independently reviewed the cine loop of the arterialphase images of CEUS examinations in all 51 patients and graded the presence and pattern of tumoral vascularity, which served as the reference standard for each tumor. Similar to the assessment of the precontrast ultrasound images, the presence of vascularity on CEUS was graded on a 5-point scale: 0, no signal in the tumor; 1, a few spotty signals in the tumor; 2, signals in <25% of the tumor; 3, signals in 25%-50% of the tumor; and 4, signals in >50% of the tumor [16]. A score of 3 or higher was considered as positive for the presence of vascularity. The pattern of vascularity was assessed as follows: basket pattern, vessels in the tumor, spot pattern, detouring pattern, a combination of above, or others (Figs. 2, 3) [17]. After independent review, the two radiologists discussed their findings and reached a consensus.

Data were represented as mean±standard deviation (SD), or as absolute numbers (%), as appropriate. The detection and characterization of tumoral vascularity were compared between CDI and MFI, and between PDI and MFI by using the Wilcoxon signed-rank test. The McNemar test was used to compare the sensitivity and specificity of CDI and MFI, and PDI and MFI in the detection and characterization of tumoral vascularity. Both the presence and pattern of vascularity were considered. The results of CDI, PDI, and MFI were regarded as positive only if the pattern of tumoral vascularity on each examination matched that of the CEUS examination. Spearman correlation coefficients (ρ) were calculated to assess the correlation between the vascular patterns observed on MFI and CEUS. Univariate logistic regression analyses were performed to identify factors associated with the detection of tumoral flow on MFI. Variables showing P<0.1 in the univariate analyses were included in a multivariate logistic regression analysis to reveal the factors independently associated with tumoral vascularity detection. Receiver operating characteristic (ROC) curve analysis was performed to obtain cutoff values for the application of MFI to assess tumoral vascularity. Lastly, interobserver agreement was assessed by using the intraclass correlation coefficient (ICC). We used the following convention to interpret the results of ICC: <0.20, poor; 0.21-0.40, fair; 0.41-0.60, moderate; 0.61-0.80, substantial; and 0.81-1.00, near-perfect [18]. P-values of <0.05 were considered to indicate statistical significance. Commercially available software products were used for statistical analysis (MedCalc version 18.6, MedCalc Software, Ostend, Belgium; IBM SPSS Statistics for Windows version 22.0, IBM Corp., Armonk, NY, USA).

Of the 59 total patients, eight patients were excluded; of these, five refused to participate in the study and three were excluded because of poor ultrasound visibility. In total, 51 patients with either pathologically confirmed HCC (n=8) or radiologically proven HCC (n=43) were enrolled. The characteristics of the patients and tumors are summarized in Table 1. On CEUS examinations, tumoral vascularity was demonstrated in 98% (50 of 51) of the patients. The patterns of vascularity on CEUS were as follows: basket pattern (n=26), vessels in tumor (n=6), a combination of the above (n=3), or others (n=15). No adverse events were associated with the CEUS examinations.

The vascularity score was significantly higher with MFI than CDI or PDI (P<0.001 for both). MFI also demonstrated a significantly higher sensitivity (58.0%, 29 of 50) than CDI (14.0%, 7 of 50) or PDI (14.0%, 7 of 50) (P<0.001 for both) (Fig. 3). The correlations between the vascular patterns depicted on MFI and CEUS are presented in Table 2 (ρ=1.000, P<0.001). Interobserver agreement for the detection and characterization of tumoral vascularity was higher in MFI (0.725) than in CDI (-0.268) or PDI (0.360) (Table 3).

On univariate logistic regression analysis, tumor depth (odds ratio [OR], 0.360 per centimeter; P=0.001) and whether the tumor had recurred or not (OR, 0.277; P=0.033) demonstrated statistical significance (Table 4). Regarding tumor depth, MFI demonstrated a sensitivity of 94.4% (17 of 18) and 87.5% (28 of 32) for the detection of tumoral vascularity in tumors with a depth ≤3 cm and ≤4 cm, respectively. In terms of recurrence, MFI detected 82.1% (23 of 28) of treatment-naive HCCs, but only 68.2% (15 of 22) of recurrent HCCs. The Youden index obtained from the ROC analysis revealed that tumor depth ≤3.7 cm yielded the largest area under the ROC curve (0.737), with a sensitivity and specificity of 73.7% and 66.7%, respectively. In terms of tumor size, ROC analysis demonstrated that a diameter >2.5 cm yielded the largest area under the ROC curve (0.836), with a sensitivity and specificity of 60.5% and 100.0%, respectively. When we performed univariate logistic regression analysis using the dichotomization for tumor depth and size obtained by applying the cutoff values from the ROC analysis, tumor depth also demonstrated a significant association with MFI performance (OR, 0.580; P=0.012), but tumor size did not (OR, 21.7; P=0.998). In the multivariate analysis, underlying liver disease and tumor size (P=0.081 and P=0.095, respectively) were also included. Multivariate logistic regression analysis revealed that tumor depth was the only factor significantly associated with detectability on MFI (OR, 0.353 per centimeter; P=0.005) (Table 4).