The institutional and regional science and research ethics committee of our university approved this single-center retrospective study. Given the retrospective nature of the study, the ethics committee waived the requirement for written informed consent from participants. However, all patients provided written consent for CEUS, contrast-enhanced (CE) CT, MRI examinations, and liver biopsies.

We retrospectively collected scans from 80 patients with 100 FLLs who underwent MVFI as part of abdominal US between June 2021 and November 2023. MVFI was not universally applied to all FLLs, but rather to selected cases in which the examiner determined it could facilitate a correct differential diagnosis. Eligibility criteria included: age 18 years or older, diagnosis or follow-up of FLLs, availability of MVFI scans of the lesions, and diagnosis confirmed by independent imaging studies or pathology report. Patients were excluded if they declined participation, had non-identifiable liver lesions on B-mode US had decompensated chronic liver disease, or the diagnosis could not be established based on imaging and histology studies.

FLL diagnoses were confirmed by dynamic contrast-enhanced imaging, or biopsy, following international recommendations and institutional guidelines. A summary of lesions by category is provided below: HEMs either had a typical hyperechoic appearance and were stable on follow-up (n = 4), or showed early-phase peripheral globular enhancement with delayed centripetal filling on CE imaging (n = 6). FNHs were confirmed by HB-MRI findings including isointensity or hypointensity on T1-weighted images, isointensity or slight hyperintensity on T2-weighted images, early enhancement, lack of venous washout, and iso- or hyperintensity in the hepatobiliary phase with or without a hypoenhancing central scar (n = 19); by CEUS showing a central feeding artery, centrifugal filling, homogeneous arterial enhancement, and no washout in venous or delayed phases (n = 15); or by biopsy (n = 13). HCAs were diagnosed by HB-MRI showing arterial hyperenhancement, isointensity in the venous phase, and hypointensity in the hepatobiliary phase, in lesions <5 cm (n = 9); or confirmed by biopsy or surgical excision (n = 8). METs were identified either by biopsy/surgical excision (n = 12), or based on the presence of multiple newly detected hypoenhancing lesions on CE-CT or MRI in patients with known malignancy (n = 19). HCCs were diagnosed in the context of chronic liver disease based on classic imaging features (early enhancement and washout in portal venous/delayed phases, with or without an enhancing capsule) (n = 21), or by biopsy (n = 20). Cirrhosis was present in five patients diagnosed with HCC (n = 7). In one cirrhotic patient, a lesion (n = 1) with arterial enhancement was diagnosed as FNH on HB-MRI. Among HCC patients, six (n = 7 lesions) had prior transarterial chemoembolization (TACE), two (n = 3) received thermal ablation, and two (n = 3) underwent systemic chemotherapy. Additionally, twelve patients with metastatic disease (n = 14 lesions) received chemotherapy before the MVFI examination.

Demographic data, medical history, and diagnostic results were collected from electronic medical records. Table 1 summarizes the characteristics of the patient cohort and the distribution of lesion types.

All patients were examined using a Samsung RS85 Prestige US system (Samsung Medison Co., Ltd., Seoul, Korea) with a CA1-7S convex probe. Scans were performed after at least 4 h of fasting, with patients positioned supine or in left lateral decubitus, arms raised above the head. All examinations were conducted by a radiologist with over 10 years of experience in abdominal US. Lesions were first identified on B-mode US. The MV-Flow™ mode was then activated to assess microvascular flow patterns. Each lesion was visualized in B-mode, after which the MV-Flow™ window was applied. Still images and 5–10 s video clips were recorded while patients held their breath. Imaging settings were optimized for patient body habitus and lesion location, with typical parameters including an average flow velocity <2 cm/s, dynamic range of 50 dB, and frame rate of 40 fps. The focus was adjusted according to lesion depth. In 27 FNH lesions, vascularity was also evaluated using directional Power Doppler imaging (PDI).

US and MVFI images were retrospectively anonymized and evaluated. Lesion size, echogenicity, margin clarity, depth from liver surface (DFS), and segmental location were assessed on B-mode images. Two independent consultant radiologists, each with more than 10 years of experience and blinded to clinical data and each other’s findings, reviewed the anonymized MVFI images. Lesions were categorized based on established MVFI flow patterns (Figure 1). Although no standardized classification scheme for evaluating MVFI patterns has been established, previous studies consistently describe a spoke-wheel pattern for FNH; non-specific hypervascular patterns for other primary liver tumors; dotted or nodular rim patterns for hemangiomas; and hypovascular or rim-type vascularity for most secondary liver lesions [9, 10, 13]. The MVFI patterns used to assess vascularity in the present study reflect these principles, are identical to those reported in an earlier publication from our research group, and have demonstrated good discriminatory ability among commonly encountered FLL types [12].

The observers also assessed the presence of a central artery, vascular rim, vascular density, and technical image quality using a three-level scale (poor, acceptable, good). Details of evaluated features are listed in Supplementary Table S1.

Quantitative variables with normal distribution are reported as mean ± standard deviation (SD); non-normally distributed variables are presented as median and range. Categorical variables are shown as frequencies and percentages. The Kruskal–Wallis test was used to compare patient ages across lesion types. Fisher’s exact test assessed the distribution of MVFI features among FLLs. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated to assess associations between MVFI patterns and lesion types, comparing FNHs to other FLLs. Inter-rater agreement was evaluated using Cohen’s kappa (ƙ). Ordinal logistic regression was applied to identify factors influencing MVFI image quality. Logistic regression models using combinations of MVFI and grayscale features were trained with leave-one-out cross-validation (LOOCV) to distinguish FNH from other FLLs. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curves, and the area under the ROC curve (AUC) was reported.

A priori power analysis was performed to ensure a type II error rate below 20% in the ROC analysis [14]. Diagnostic metrics: sensitivity (Se), specificity (Sp), positive predictive value (PPV), and negative predictive value (NPV) were calculated for all lesions, as well as separately for lesions <3 cm and ≥3 cm in diameter. Model comparisons were made using the likelihood ratio test.

Statistical significance was set at p < 0.05. The Dunn-Sidak correction was applied for multiple comparisons [15]. All analyses were performed using R software (version 4.3.2; ¹ , accessed 01/12/2023).

The primary objective of this retrospective case–control study was to evaluate MVFI patterns capable of distinguishing FNH from other common FLLs. To achieve this, we aimed to include a substantial number of FNH cases for a comprehensive assessment of their vascular characteristics, along with a control group representing the typical differential diagnoses of FNH, thereby enabling us to assess the diagnostic potential of MVFI. In total, we enrolled 80 patients with 100 FLLs. Because this was a retrospective case–control design focused specifically on comparing FNH with other focal liver lesions, the distribution of lesion types in our sample does not reflect their true incidence in clinical practice. Consequently, FNH constituted the largest subgroup in our cohort (n = 41). It was diagnosed in 31 patients, of which 24 were females with a median age of 44.5 years (range: 20–66 years), and 7 were males with a median age of 40 years (range: 22–68 years). Three patients had three, four patients had two, and twenty-four patients had a single lesion. The second most common FLL was HCC (n = 21), which was diagnosed in 18 patients, including four females with a median age of 77.5 years (range: 65–83 years) and 14 males with a median age of 69 years (range: 56–79 years). HCAs (n = 10) were present in six females with a median age of 41 years (range: 26–45 years). The study cohort also included 17 patients with METs (n = 20) originating from various types of primary tumors, and eight patients diagnosed with HEM (n = 8).

As expected, patients diagnosed with FNH or HCA were significantly younger than those with hepatocellular carcinoma HCC or MET, with all pairwise comparisons yielding p-values <0.012. Male sex was a significant risk factor for HCC, with an OR of 0.12 (95% CI: 0.03–0.35; p < 0.0001), while HCA was observed exclusively in female patients within our cohort.

Hepatic steatosis was identified in 18 patients (22.5%), who collectively harbored 26 FLLs. Background liver cirrhosis was present in 11 patients and was associated with seven HCCs, three METs, two HEMs, and one case of FNH. Prior to MVFI, 22 patients had received locoregional or systemic treatment, involving 13 cases of HCC and 14 METs.

We recorded lesion size, anatomical location, echogenicity, margin definition, DFS for all FLLs. The median size of FNH lesions was 38 mm (range: 12–83 mm), and the median DFS was 34 mm (range: 12–88 mm), with no statistically significant differences compared to other FLL types. FNH lesions were most commonly isoechoic relative to the surrounding liver parenchyma (n = 25; 51%), followed by hypoechoic (n = 11; 27%) and slightly hyperechoic (n = 5; 12%) appearances. The majority of FNHs were located in the right hepatic lobe (n = 25; 61%) and had a peripheral distribution (n = 27; 66%).

Compared to other FLL types, FNH lesions were significantly more likely to appear isoechoic (OR = 11.61; 95% CI: 4.43–33.92; p < 0.0001) and to have poorly defined margins (OR = 4.17; 95% CI: 1.72–10.00; p = 0.0014). A central scar was identified in 4 FNH lesions, accounting for 10% of cases.

The spoke-wheel vascular pattern was the characteristic MVFI feature of FNH. It was identified in 35 FNH lesions (85%) by the first observer and in 26 lesions (88%) by the second observer (Figure 2). A non-specific hypervascular pattern was observed in 4 FNHs (10%) by the first observer and in 3 FNHs (7%) by the second. Alternative MVFI patterns, including spotted-dot, nodular rim with central dot, and basket-wave configurations, were each observed in only one FNH lesion (2%) by either observer. Among FNH lesions measuring <3 cm, the spoke-wheel pattern was detected in 12 cases (80%) by both observers. The spoke-wheel pattern was highly specific for FNH, as indicated by the observations of the first (OR = 51.53; 95% CI: 16.68–191.51; p < 0.001) and second observer (OR = 35.28; 95% CI: 12.00–124.65; p < 0.001).

However, spoke-wheel-like appearances were occasionally identified in non-FNH lesions: the first observer noted this pattern in 3 HCCs (14%), 2 HCAs (20%), and 1 MET (5%), the second observer in 4 HCCs (19%), 3 HCAs (30%), and 3 METs (15%).

The presence of a central artery on MVFI was found to be a characteristic feature of FNH according to the first observer (OR = 4.96; 95% CI: 1.19–23.00; p < 0.03), but this association was not statistically significant for the second observer (OR = 1.85; 95% CI: 0.45–7.10; p = 0.373). The first observer identified a central artery in 30 FNH lesions (73%), as well as in 4 HCCs (19%), 3 METs (15%), and 2 HCAs (20%) (Figure 3). In comparison, the second observer detected a central artery in 32 FNHs (78%), 8 HCCs (38%), 6 METs (30%), and 3 HCAs (30%).

Hypovascular lesions, including those with sparse or no MVFI signal, accounted for 0% (0/41) of FNHs and 44% (26/59) of other FLLs according to the first observer, and 2% (1/41) of FNHs and 8% (5/59) of other FLLs according to the second observer.

We assessed the interobserver reliability of MVFI features between the two observers. Substantial agreement was observed for the six predefined MVFI vascular patterns (κ = 0.641; p < 0.001) (Figure 4). The reproducibility of the spoke-wheel pattern was even higher, with a κ value of 0.748 (p < 0.001). Agreement was also substantial for the presence of a central artery (κ = 0.638; p < 0.001), while interobserver consistency for rim vascularity was moderate (κ = 0.475; p < 0.001).

Image quality was rated by both observers using a three-tier scale. According to the first observer’s evaluation, increasing DFS was significantly associated with poorer image quality (OR = 1.64; 95% confidence interval CI: 1.30–2.11; p < 0.001). Additionally, lesions located in the left hepatic lobe were more likely to receive lower quality scores (OR = 2.00; 95% CI: 1.05–8.52; p = 0.045). In contrast, the second observer did not find a statistically significant association between image quality and DFS (OR = 1.22; 95% CI: 0.98–1.53; p = 0.077) or lesion localization (OR = 0.72; 95% CI: 0.25–1.96; p = 0.552). Lesion size and central versus peripheral location were not significantly associated with image quality in the assessments of either observer.

Data from both observers were combined into a single dataset, and logistic regression models were developed using LOOCV to differentiate FNH from other types of FLLs (Table 2). A model incorporating MVFI features, specifically, the presence or absence of a spoke-wheel pattern and central artery, demonstrated significantly higher diagnostic accuracy (AUC = 0.891; p < 0.001) compared to a model based solely on grayscale US features, such as echogenicity and margin definition (AUC = 0.814).

The MVFI-based model misclassified 6 FNH lesions (15%), 3 HCCs (14%), 2 HCAs (20%), and 1 MET (5%) according to the first observer’s evaluation. Based on the second observer’s assessment, the model misclassified 5 FNHs (12%), 4 HCCs (19%), 3 HCAs (30%), and 3 METs (15%) (Supplementary Table S2).

Notably, the MVFI-based model showed even greater diagnostic accuracy (AUC = 0.944, p < 0.001) and consistently outperformed the classic grayscale model (AUC = 0.879) for FNH lesions measuring ≥3 cm.