Preliminary study on the ability of ¹⁸F-fluorodeoxyglucose positron emission tomography/computed tomography radiomics to predict vessels that encapsulate tumor clusters and prognosis in hepatocellular carcinoma

Introduction

Hepatocellular carcinoma (HCC) is the sixth most common cancer worldwide and the third leading cause of cancer-related death (1), imposing a major threat and burden to public health. Despite advances and improvements in the treatment of patients over the last decades, the prognosis of HCC patients following radical surgery remains unfavorable due to the high risk of early recurrence and metastasis (2,3).

HCC is a typical hypervascular neoplasm with abundant angiogenesis and a malformed vascular network, rendering it susceptible to early hematogenous dissemination (4) and resulting in an unfavorable prognosis (5,6). Recently, a novel vascular pattern characterized by vessels encapsulating tumor clusters (VETC) was reported to be related to poor clinical outcomes in HCC patients (7,8). The unique structure of VETC is independent of the malignant tumor metastasis process based on the epithelial-mesenchymal transformation (EMT) theory, providing a new pattern for metastasis (9). Notably, VETC-positive HCC exhibits higher metastasis efficiency (7), so this indicator has also emerged as an independent predictor of recurrence and overall survival (OS) in patients undergoing hepatectomy (10,11) and liver transplantation (LT) (12,13). Furthermore, VETC status can serve as a predictive indicator for response to other therapies such as trans-arterial chemoembolization and sorafenib, providing valuable guidance for treatment decision-making (14,15).

However, the preoperative identification of VETC pattern in HCC remains challenging due to its reliance on pathological slides for diagnosis. Several studies have investigated image-based noninvasive methods, such as ultrasound (US) (16), computed tomography (CT) (11), and magnetic resonance imaging (MRI) (17-19), to identify VETC-positive HCCs before treatment, with promising results. Nevertheless, the intrinsic link between tumor morphological characteristics and VETC expression remains questionable. The study conducted by Itoh et al. (20) revealed a strong correlation between glucose metabolism levels and angiogenesis of VETC, suggesting that the metabolic characteristics observed on ¹⁸F-fluorodeoxyglucose positron emission tomography/computed tomography (FDG PET/CT) may hold predictive value for preoperative assessment of VETC status. Radiomics, an emerging method in medical image analysis, employs multidimensional mathematical calculation to comprehensively extract and analyze image information with the aim of further supporting decision-making and improving clinical diagnostic efficiency (21). Radiomics has been widely applied in HCC, encompassing differential diagnosis, therapeutic response evaluation, and early recurrence prediction (22-24). To the best of our knowledge, there is currently a lack of studies investigating the potential of ¹⁸F-FDG PET radiomics features in predicting VETC and patient prognosis in HCC.

The purpose of this study was to assess the efficacy of ¹⁸F-FDG PET metabolic parameters in distinguishing VETC status compared to traditional CT/MRI image features, and investigate the predictive capability of an ¹⁸F-FDG PET radiomics model in preoperative identification of VETC-positive HCCs. We present this article in accordance with the TRIPOD+AI reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2734/rc).

Methods

Patients

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The Third Affiliated Hospital of Sun Yat-sen University (No. II2024-113-01). Both participating institutions were informed of and agreed to the study. Informed consent was provided by all the patients.

This study retrospectively analyzed patients who underwent ¹⁸F-FDG PET/CT examination for suspected primary liver cancer at Institution 1 (The Third Affiliated Hospital of Sun Yat-sen University) between June 2015 and February 2022 as the training cohort and at Institution 2 (Sun Yat-sen Memorial Hospital) between June 2017 and February 2022 as the test cohort. The inclusion criteria for patients were as follows: (I) underwent radical surgical treatment (hepatectomy or LT); (II) pathologically confirmed HCC; (III) HCC underwent CD34 immunochemical staining; (IV) ¹⁸F-FDG PET/CT was performed within 1 month before surgery, upper abdominal CT and/or MRI were performed during the same period. The exclusion criteria were as follows: (I) preoperative ¹⁸F-FDG PET/CT, upper abdominal CT or MRI images were of poor quality; (II) underwent other cancer treatments before surgery or had history of other malignant tumors; and (III) incomplete clinical data. The flowchart of patient selection is shown in Figure 1.

Figure 1 The flowchart of patient selection. Institution 1: The Third Affiliated Hospital of Sun Yat-sen University; Institution 2: Sun Yat-sen Memorial Hospital. CT, computed tomography; FDG PET/CT, fluorodeoxyglucose positron emission tomography/computed tomography; HCC, hepatocellular carcinoma; MRI, magnetic resonance imaging; VETC, vessels that encapsulate tumor cluster.

Preoperative clinical and laboratory characteristics were collected including age, sex, hepatitis B virus (HBV) infection, alpha-fetoprotein (AFP), aspartate transaminase (AST), alanine aminotransferase (ALT), Child-Pugh, and Barcelona Clinic Liver Cancer (BCLC) stage. Pathological data included cirrhosis, differentiation, microvascular invasion (MVI), Ki-67, and VETC.

Histopathology and immunohistochemistry

VETC was identified on CD34 immunochemical staining. All surgical specimens were reviewed by a board-certified pathologist with over 7 years of specialized experience in liver pathology. The VETC pattern was regarded as the presence of sinusoid-like vessels that formed cobweb-like networks and encapsulated individual tumor clusters with an unequivocal and continuous lining of CD34-positive endothelium in the whole or part of the tumor (≥5% tumor area) (7).

PET/CT examination

PET/CT was performed using a Discovery Elite scanner (Institution 1: GE Healthcare, Milwaukee, WI, USA) or a Biograph 64 PET/CT scanner (Institution 2: Siemens Healthcare, Erlangen, Germany). The patients fasted for at least 6 hours, and glucose levels in the peripheral blood were confirmed to be lower than 200 mg/dL (11.1 mmol/L) before FDG injection. An unenhanced CT scan (120 kV, 120 mA and slice thickness 3.75 mm) was performed at 60 minutes after intravenous injection of 3.70–5.55 MBq/kg of ¹⁸F-FDG. Subsequently, whole-body PET imaging was acquired, ranging from the skull to the thighs, with acquisition times of 2 minutes per bed position (7–9 beds for Discovery Elite) or 1.2–1.5 minutes per bed position (6–8 beds for Biograph 64). The PET images were reconstructed using the ordered-subset expectation maximization (OSEM) algorithm with attenuation correction. Additionally, the VUE Point FX module of Discovery Elite system and a TrueX + TOT algorithm of Biograph 64 system were employed for further processing.

Traditional image features

Traditional image features obtained by upper abdominal enhanced CT or MRI included the following: (I) intratumor necrosis; (II) non-smooth tumor margin; (III) peritumoral arterial enhancement; and (IV) intratumor arteries. Traditional image features obtained by PET/CT included: (I) tumor size; (II) tumor number; (III) tumor-to-liver ratio (TLR): calculated as the maximum standardized uptake value (SUVmax) of the tumor/the mean standardized uptake value (SUVmean) of the normal liver. The details of evaluation rules of traditional image features are summarized in Appendix 1. The correlation between traditional image features and VETC status was estimated. All of these image features were evaluated by two radiologists with 6 and 14 years of experience in abdominal imaging, who were blinded to clinical and pathologic results of the patients. Occasional inconsistencies were discussed to reach a consensus.

PET/CT radiomics analysis

The tumor regions of interest (ROIs) for each patient were manually delineated slice-by-slice by two experienced radiologists, with 3 and 5 years of professional experience, respectively, using the 3D Slicer software (version 4.10.2; http://www.slicer.org/). Both radiologists were blinded to the clinical outcomes of the patients. Lesions were defined as areas exhibiting abnormal uptake of ¹⁸F-FDG on PET images and/or abnormal density on CT images. The ROIs were drawn section-by-section to measure the corresponding volumes of interest (VOIs) on both PET and CT images, respectively. In cases where the lesion demonstrated low uptake of ¹⁸F-FDG, the ROI was initially determined on CT images and subsequently transferred to PET images to define the corresponding PET ROI.

Preprocessing was performed before feature extraction, including image resampling and grayscale normalization. The images were resampled at a spatial resolution of 1×1×1 mm³ using BSpline interpolation and discretized with a bin width of 25. Radiomics features were extracted from both PET and CT images by the pyradiomics package (https://pyradiomics.readthedocs.io/). The feature pool of each sequence comprised 14 shape features, 18 first-order statistical features, and 75 texture features. All the radiomics features are summarized in Table S1. The intraclass correlation coefficient (ICC) was used to evaluate the interobserver reproducibility of radiomics feature extraction using images of 30 randomly selected patients. ICC, a statistical measure ranging from 0 to 1, indicates the reliability of measurements, with 0 representing unreliability and 1 indicating complete reliability. In this study, only features with ICC values exceeding 0.75 were considered beneficial for subsequent research.

Normalization was applied to each radiomics feature to eliminate the influence caused by numerical range differences among features. The least absolute shrinkage and selection operator (LASSO) regression algorithm was employed to identify the optimal set of contributing features. Specifically, features with non-zero coefficients in the LASSO regression results were retained. Subsequently, a radiomics score (Radscore) was calculated for each patient as a linear combination of selected features weighted by their corresponding coefficients. The detailed radiomics workflow is shown in Figure 2.

Figure 2 The workflow of radiomics analysis. FDG PET/CT, fluorodeoxyglucose positron emission tomography/computed tomography.

Model development and validation

In the training cohort, clinical characteristics, traditional PET image features, and Radscore were investigated for their association with VETC by logistics regression analysis. Indicators that demonstrated statistical significance (P<0.05) in the univariate analysis were subsequently subjected to further evaluation using multivariate logistic regression. The nomogram model was created by all independent factors including Radscore. The clinical model was created by clinical characteristics and traditional PET image features. The area under the curve (AUC) was utilized as a metric to assess the predictive performance of the nomogram model, clinical model, Radscore, and TLR for prediction of VETC-positive status. Calibration curves were plotted to assess the calibration of the nomogram model and clinical model. To estimate the clinical utility of different models, we performed decision curve analysis (DCA) with decision curves based on the nomogram model, clinical model, Radscore, and TLR to calculate the net benefit for the whole cohort at a range of threshold probabilities.

Follow-up

Patients in the training cohort underwent regular follow-up and telephone follow-up until 28 February 2023. The review protocol was performed every 3–6 months after radical surgery for AFP assessment and imaging examination (US, CT, or MRI) to determine local recurrence or distant metastases. The primary end points were disease-free survival (DFS) and OS. DFS was defined as the time from the date of surgery to the date of HCC recurrence or death or the date of last follow-up. OS was defined as the time from the date of surgery to the date of death or the date of last follow-up.

Statistical analysis

Continuous variables that conformed to a normal distribution were analyzed using a t-test and represented by the mean ± standard deviation (SD), whereas variables not conforming to a normal distribution were compared using Mann-Whitney U test and represented by the median with the interquartile range (IQR). Categorical variables were analyzed using chi-squared test or Fisher’s exact test and reported by numbers with percentages. The AUC was plotted to evaluate the predictive ability of different variables and models. The Delong test was used to compare the AUC values. The optimal cutoff value was defined by Youden index. Survival curves were analyzed using the Kaplan-Meier method and compared using the log-rank test. All statistical analyses were performed using the software SPSS 25.0 (IBM Corp., Armonk, NY, USA) and R version 4.1.3 (http://www.R-project.org). Differences were considered significant when a two-tailed P value was <0.05.

Results

Baseline characteristics

The clinical and pathological characteristics of HCC patients in two institutions are displayed in Table 1. A total of 103 patients (91 male and 12 female; 53.0±11.4 years) were included in the training cohort, and 46 patients (41 male and 5 female; 55.9±11.2 years) were included in the test cohort. VETC positivity was detected in 53.4% (55/103) of tumors in the training cohort, which was similar to the 50.0% (23/46) seen in the test cohort (P=0.701). The comparison of pathological characteristics showed that there were more patients in the training cohort who had MVI (P=0.007) and tumor differentiation of moderate stage (P=0.006), whereas no differences were found in other characteristics between the two cohorts. In the training cohort, patients with VETC-positive HCC exhibited significantly elevated AFP levels (P=0.001), AST levels (P<0.001), Ki67 expression (P=0.001), and more MVI (P=0.005) compared to those with VETC-negative HCCs. There were no statistically significant differences in other characteristics between VETC-positive and VETC-negative patients in the training cohort (P>0.05).

Traditional image features with VETC status

Table 2 details the traditional image features based on CT/MRI or PET/CT in the two cohorts. VETC-positive HCCs exhibited a higher prevalence of intratumor necrosis, peritumoral arterial enhancement, intratumor arteries, tumor size greater than 5 cm (all P<0.05) compared to VETC-negative HCCs in the training cohort. Patients with VETC-positive HCC exhibited significantly higher TLR levels compared to those with VETC-negative HCC [median, 1.59 vs. 2.45; IQR, 1.36–2.02 vs. 1.98–3.41, P<0.001]. Representative images of traditional image features are shown in Figure 3.

Figure 3 Typical manifestations of traditional image features. (A) A 65-year-old man with VETC-positive, contrast-enhanced CT scan in arterial phase shows lesion located in the right liver lobe; yellow arrow indicates intratumor necrosis, red arrows show multiple tortuous and thickened intratumor arteries, the irregular enhancement area at the edge of the lesion suggests peritumoral enhancement (green arrows). (B) A 65-year-old man with VETC-negative, MRI shows the lesion located in the right liver lobe, the white arrows indicate the non-smooth tumor margin. (C) A 51-year-old man with VETC-positive, fused PET/CT shows lesion located in the right liver lobe. The lesion shows significantly increased FDG uptake, with a TLR of 3.10. FDG, fluorodeoxyglucose; MRI, magnetic resonance imaging; PET/CT, positron emission tomography/computed tomography; TLR, tumor-to-liver ratio; VETC, vessels encapsulating tumor clusters.

According to receiver operating characteristic (ROC) analysis, a TLR of 2.06 was the best cutoff value to distinguish VETC status, so HCC patients were divided into TLR ≤2.06 and TLR >2.06 groups. The performance of intratumor necrosis, peritumoral arterial enhancement, intratumor arteries, tumor size >5 cm and TLR >2.06 in predicting VETC status as independent predictors is shown in Table 3. Compared with traditional CT/MRI image features, PET/CT metabolic parameter TLR possessed higher AUC values [0.789, 95% confidence interval (CI): 0.698–0.880 vs. 0.684, 95% CI: 0.581–0.787] and relatively superior sensitivity and specificity in the training cohort, while demonstrating comparable predictive performance in the test cohort. The Delong test revealed statistical differences in the AUC values between TLR >2.06 and other image features in the training cohort (Table S2). These results demonstrated a significant association between TLR and VETC status, which held predictive value for VETC prior to surgery when compared to traditional CT/MRI image features.

PET/CT radiomics with VETC status

Potential radiomics features were selected to establish Radscore using the LASSO regression algorithm based on 103 patients in the training cohort. LASSO analysis revealed that six PET radiomics features (kurtosis, skewness, mean, elongation, GLDM_dependence non-uniformity normalized, maximum 2D diameter row) and seven CT radiomics features (90th percentile, maximum, mean absolute deviation, flatness, GLRLM_run variance, GLSZM_small area emphasis, NGTDM_contrast) were most useful for predicting VETC status (Figure S1). The selection of radiomics features using LASSO regression is shown in Figure S2. The calculation formula for Radscore based on selected Radscore is presented in Appendix 1. There was a significant difference in Radscore between VETC-positive HCCs (training cohort: median, 1.27; IQR, −0.28, 2.173; test cohort: median, 0.712; IQR, −1.464, 1.799) and VETC-negative (training cohort: median, −0.96; IQR, −2.20, 0.211; test cohort: median, −0.97; IQR, −2.650, 0.182) HCCs in the training cohort (P<0.001) and test cohort (P=0.008) (Figure S3).

Models for preoperatively predicting VETC status

In univariate analysis, significant differences were observed in AFP >400 ng/mL, AST >40 U/L, TLR, tumor size >5 cm and PET Radscore between VETC-positive and VETC-negative HCCs. Subsequently, multivariable logistic regression analysis confirmed that AFP >400 ng/mL [odds ratio (OR) =4.18, P=0.029], AST >40 U/L (OR =4.33, P=0.015), TLR >2.06 (OR =4.03, P=0.028), and Radscore (OR =2.09, P=0.003) were independent predictors of VETC status (Table 4).

A nomogram model for preoperative prediction of VETC status was established using the above four independent risk factors. The nomogram based on the regression coefficients was plotted from these factors and shown in Figure 4A. AFP >400 ng/mL, AST >40 U/L and TLR >2.06 were combined to construct a clinical model. Compared with the clinical model, Radscore, and TLR >2.06 (Figure 4B,4C), the nomogram model obtained an AUC of 0.908 (95% CI: 0.852–0.963) in the training cohort, and an AUC of 0.762 (95% CI: 0.624–0.900) in the test cohort (Table 5). The AUCs of the nomogram were higher than that of the clinical model, Radscore, and TLR >2.06 (Table 5), although there were no statistical differences in the AUC values among most of the groups (Table S3). The calibration curves for predicting VETC status of the nomogram model and clinical model are shown in Figure S4A-S4D. The calibration curves demonstrated a consistently high level of agreement between the estimated values and actual observation within the training cohort. DCA showed that for the differentiation of VETC-positive from VETC-negative, the nomogram model added more net benefits than the other three models in the training cohort (Figure 4D,4E). Representative cases to show discriminative ability of nomogram model based on ¹⁸F-FDG PET/CT for identifying VETC status are shown in Figure 5.

Figure 4 Illustration of the nomogram model and demonstration of the diagnostic performance of each model. (A) Development and performance of the nomogram model. ROC curves based on four models for estimating VETC status of HCCs in the training (B) and test (C) cohort. DCA for four models in the training (D) and test (E) cohort. AFP, alpha-fetoprotein; AST, aspartate transaminase; DCA, decision curve analysis; HCC, hepatocellular carcinoma; Radscore, radiomics score; ROC, receiver operating characteristic; TLR, tumor-to-liver ratio; VETC, vessels encapsulating tumor clusters.

Figure 5 Representative cases to show discriminative ability of nomogram model based on ¹⁸F-FDG PET/CT for identifying VETC status. (A-C) A 74-year-old man with AFP >400 ng/mL, AST >40 U/L, TLR >2.06, Radscore =3.5664. (A) On CT scan, the lesion shows a slightly low-density mass (yellow arrows), located in the right liver lobe. (B) PET and (C) fused PET/CT showed the lesion was significantly hypermetabolic; according to nomogram model, the probability of VETC-positive in this patient was 0.99, which was confirmed by histological result to be VETC-positive. (D) Representative performance of VETC-positive (hematoxylin-eosin stain, ×100), presence of sinusoid-like vessels that formed cobweblike networks and encapsulated individual tumor clusters with an unequivocal and continuous lining of CD34 positive endothelium. (E-G) A 41-year-old man with AFP ≤400 ng/mL, AST ≤40 U/L, TLR ≤2.06, Radscore =−0.7566. (E) CT scan shows the lesion was a low-density shadow (yellow arrows) with poorly defined boundaries, located in the S6 segment protruding from the liver surface. (F) PET and (G) fused PET/CT shows the metabolism of the lesion was slightly increased (yellow arrows); according to nomogram model, the probability of VETC positivity in this patient was only 0.09, which was confirmed by histological result to be VETC-negative. (H) Representative performance of VETC-negative (hematoxylin-eosin stain, ×100). AFP, alpha-fetoprotein; AST, aspartate transaminase; FDG PET/CT, fluorodeoxyglucose positron emission tomography/computed tomography; Radscore, radiomics score; TLR, tumor-to-liver ratio; VETC, vessels encapsulating tumor clusters.

Survival prediction and VETC status

As of February 2023, a total of 94 (91.3%) patients had completed follow-up in the training cohort. The median duration of follow-up was 26.5 (IQR, 15.0–44.0) months. The recurrence rate was 41.5% (39/94) and the mortality rate was 30.9% (29/94). Patients with histologic VETC-positive HCC were more likely to experience recurrence than those with histologic VETC-negative HCC (P=0.0018) (Figure 6A). Similarly, when using the nomogram model, Kaplan-Meier analysis demonstrated that predicted VETC-positive HCCs exhibited significantly poorer DFS (P<0.001) (Figure 6B) compared to those with predicted VETC-negative. Moreover, whether histologic VETC-positive or predicted VETC-positive, both categories exhibited poorer OS compared to cases that were histologic VETC-negative (P<0.001) (Figure 6C) or predicted VETC-negative (P<0.001) (Figure 6D).

Figure 6 Kaplan-Meier analysis according to histological and nomogram model predicted VETC status. Comparison between histological VETC-positive and VETC-negative for DFS (A) and OS (C). Comparison between predicted VETC-positive and VETC-negative by nomogram model for DFS (B) and OS (D). DFS, disease-free survival; OS, overall survival; VETC, vessels encapsulating tumor clusters.

Discussion

VETC in HCC represents an aggressive form associated with an unfavorable prognosis. Preoperative characterization of this unique subtype may facilitate the development of individualized therapeutic strategies and improve prognostic assessment. In this study, we have demonstrated a significant association between the TLR in PET/CT images and VETC status. Moreover, our further radiomics analysis of PET/CT images revealed that the Radscore exhibited superior predictive value for determining VETC status compared to the quantitative parameter TLR. AFP level, AST level, TLR, and Radscore were shown to be independent predictors for VTEC-positive HCCs. We further developed and validated a nomogram model incorporating these features for prediction of VETC status. Our results revealed that the nomogram model had excellent predictive performance and successfully stratified patients into VETC-positive and VETC-negative groups, showing significant differences in DFS and OS, which aligns with the outcomes observed in VETC-positive and VETC-negative cases through pathological analyses. The constructed model based on ¹⁸F-FDG PET/CT radiomics emerged as a valuable tool for preoperatively predicting VETC status and prognosis after radical treatment of HCCs, providing crucial information for medical decision-making support.

The reported prevalence of VETC positivity in HCC tends to vary among different study populations and due to variations in definitions, ranging from 14.2% to 56.3% (mostly 30–40%) (25). In our primary cohorts, the prevalence of VETC positivity (53.4%, 50%) was higher than most reported studies. The definition of VETC-positive in our study adhered to the methods employed by several researchers (8,26,27), whereby a VETC pattern exceeding 5% of the entire tumor was considered positive. In the literature that defines any VETC pattern as VETC-positive, the reported incidence of VETC can vary from 55.5% (28) to 56.3% (29), or even as high as 76.5% (12). Meanwhile, the prevalence of VETC appears to broadly correlate with tumor stage and aggressiveness. The utilization of ¹⁸F-FDG PET/CT scanning as an effective restaging tool for patients with clinically suspected tumors of higher malignancy may account for the relatively high prevalence of VETC positivity observed in this study.

The findings of this study suggested that AFP >400 ng/mL (P=0.002) and AST >40 U/L (P=0.049) were independent risk factors for preoperative prediction of VETC-positive status, which was consistent with the results reported by other authors (8,19,29). AFP not only serves as a specific tumor marker for diagnosing HCC but also correlates with the degree of tumor differentiation and the patient’s survival prognosis (30,31). Furthermore, the elevated expression level of AFP observed in patients with VETC-positive HCC consistently aligns with the impact of VETC status on patient prognosis. Although there is no correlation between AST levels and the diagnosis of HCC, elevated AST levels typically indicate hepatocellular organelle damage. These increased levels may be associated with the degree of invasion in HCC patients, making them more prevalent in VETC-positive patients compared to VETC-negative cases.

Several studies have indicated a correlation between qualitative indicators observed in CT/MRI images and VETC status (11,18,19). The findings of this study were in line with previous studies, demonstrating significant differences in VETC-positive and VETC-negative status involving intratumor necrosis, peritumoral arterial enhancement, and intratumor arteries. Angiogenesis activation is a characteristic feature of VETC pattern (11,32). Neovascularity mainly emerges in the periphery of the tumor, and as the tumor expands, the diffusion distance from the existing blood vessel supply increases, consequently promoting tumor cell proliferation and leading to hypoxia (33). The concurrent occurrence of neovascularization and hypoxia contributes to the necrosis observed in rapidly growing HCCs (34). These pathophysiological processes induced by the VETC pattern manifest as intratumor necrosis, peritumoral arterial enhancement, and intratumor arteries on CT/MRI images. Additionally, there was a significant correlation between the traditional PET metabolic parameter TLR and VETC status, which was consistent with the findings reported by Itoh et al. (20). As an independent predictor of VETC status, TLR >2.06 exhibited superior predictive performance and stronger correlation compared to traditional CT/MRI image features, thereby highlighting the potential of ¹⁸F-FDG PET/CT as a valuable adjunctive tool for distinguishing VETC status. However, there can be variations in qualitative or semi-quantitative parameters across different centers due to differences in image acquisition methods. Further, the diagnostic efficacy of these parameters could be constrained when working with small sample sizes. As a simple quantitative indicator, the ability of TLR to predict VETC status requires further enhancement.

Radiomics refers to the comprehensive quantification of tumor phenotypes through the extraction of high-throughput quantitative image features. By capturing the potential spatial variability and heterogeneity of voxel intensity within the tumor, radiomics analysis can reflect changes at the cellular and genetic levels within the lesion, thereby providing more detailed information about the tumor microenvironment (35). The parameter kurtosis, for example, is a measure of the “peakedness” of the distribution of values in the image ROI, and skewness measures the asymmetry of the distribution of values about the mean value. The heterogeneity of the tumor is manifested by these variations in metabolic activity within the lesion. The significance of these two parameters has been demonstrated in the prognosis of numerous solid tumors (36-38). These parameters are also crucial radiomics features screened in this study, which can be utilized to establish a Radscore. In the present study, we conducted a comprehensive analysis of ¹⁸F-FDG PET/CT images and developed radiomics nomogram for the first time to preoperatively predict VETC status in HCC patients. The constructed ¹⁸F-FDG PET/CT Radscore exhibited superior predictive performance compared to TLR, particularly in the test cohort (AUC: 0.730 vs. 0.565). Furthermore, our established nomogram model based on PET/CT Radscores demonstrated comparable predictive performance than previously reported models (AUC: 0.908 vs. 0.812–0.897) (18,19,39,40). When compared with reported contrast-enhanced CT model (AUC training: 0.825, 95% CI: 0.747–0.903; AUC external test: 0.680, 95% CI: 0.498–0.862), our study demonstrated that the Radscore model (AUC training: 0.857, 95% CI: 0.788–0.927; AUC test: 0.730, 95% CI: 0.581–0.878) exhibited relatively superior performance in predicting VETC status. In general, the CT scan component of PET/CT may be considered inferior to contrast-enhanced CT in image description due to its specific low-dose image acquisition method without enhancement. Therefore, the improved performance of the PET/CT Radscore model in this study may be attributed precisely to the addition of PET images depicting metabolic differences.

The results of this study demonstrated a strong correlation between VETC positivity and the presence of MVI (P=0.005) as well as high Ki67 expression (P=0.001), which was consistent with findings from previously published studies (8,11). The presence of MVI aligns with the metastasis in the form of tumor cell clusters under VETC pattern, confirming the high invasiveness and efficient metastasis through blood vessels of VETC. Additionally, VETC-positive HCCs showed high expression of Ki67, suggesting the active proliferation of tumor cells and indicating a high degree of malignancy and aggressiveness in VETC-positive tumors. In summary, VETC represents a highly aggressive subtype of HCC, as numerous studies have reported its association with poor prognosis (14,27,41,42), and similar results were obtained in our study. VETC risk stratification based on histological results and the constructed nomogram significantly correlated with recurrence (histological VETC status: P=0.0018, predicted VETC status: P<0.001) and OS (histological VETC status: P<0.001, predicted VETC status: P<0.001) following radical HCC resection. These results suggest that the constructed nomogram model can offer guidance for individualized treatment and prognosis assessment in patients with HCC.

There are several limitations to this study. First, the retrospective nature of the study may have rendered it susceptible to selection biases despite external testing being conducted to improve reliability. The findings of this study may not fully represent the entire clinical spectrum, as they are limited to surgically resected HCC lesions that exhibited CD34 immunochemical staining. Second, the present study provides a preliminary investigation into the disparity between PET image features and CT/MRI image features as independent indicators to predict VETC status. Further studies should respectively compare the efficacy of different imaging methods (CT, MRI, and PET) and their radiomics models in preoperative prediction of VETC, providing a basis for the selection of clinical guidance examination methods. Additionally, as a relevant marker of neovascularization, the novel MRI biomarker diffusion-derived vessel density (DDVD) (43) may serve as a robust indicator for preoperative diagnosis of VETC, thereby offering a novel perspective for future research. Third, although this study employed a two-center design and was validated in external data, the relatively small sample size in both training and test cohorts may compromise the model’s generalization ability and affect its sensitivity and specificity. A multicenter prospective study should be applied in further research to validate the ability of the results of this study. Finally, our analysis only utilized the basic radiomics analysis method for preliminary research, and incorporating higher-order radiomics features and advanced machine learning methods may potentially enhance model performance.

Conclusions

The metabolic activity in ¹⁸F-FDG PET/CT images exhibited a significant correlation with VETC status. We developed a nomogram model for prediction of VETC status and prognosis in HCC by merging PET/CT Radscore with clinical risk factors. The radiomics nomogram can serve as a convenient tool to enhance the preoperative prediction of VETC status and prognosis, including OS and DFS, in HCC patients, which might serve as a potential, noninvasive, and effective complement to assist in formulating postoperative surveillance plans.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD+AI reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2734/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2734/dss

Funding: This work was supported by Open Research Fund Program of Collaborative Innovation Center for Molecular Imaging of Precision Medicine (grant No. 2020-ZD03).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2734/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study was approved by the Ethics Committee of The Third Affiliated Hospital of Sun Yat-sen University (No. II2024-113-01). Both participating institutions were informed of and agreed to the study. Informed consent was provided by all the included patients.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
2. Forner A, Reig M, Bruix J. Hepatocellular carcinoma. Lancet 2018;391:1301-14. [Crossref] [PubMed]
3. Vogel A, Meyer T, Sapisochin G, Salem R, Saborowski A. Hepatocellular carcinoma. Lancet 2022;400:1345-62. [Crossref] [PubMed]
4. Llovet JM, Kelley RK, Villanueva A, Singal AG, Pikarsky E, Roayaie S, Lencioni R, Koike K, Zucman-Rossi J, Finn RS. Hepatocellular carcinoma. Nat Rev Dis Primers 2021;7:6. [Crossref] [PubMed]
5. Konyn P, Ahmed A, Kim D. Current epidemiology in hepatocellular carcinoma. Expert Rev Gastroenterol Hepatol 2021;15:1295-307. [Crossref] [PubMed]
6. Yao C, Wu S, Kong J, Sun Y, Bai Y, Zhu R, Li Z, Sun W, Zheng L. Angiogenesis in hepatocellular carcinoma: mechanisms and anti-angiogenic therapies. Cancer Biol Med 2023;20:25-43. [Crossref] [PubMed]
7. Fang JH, Zhou HC, Zhang C, Shang LR, Zhang L, Xu J, Zheng L, Yuan Y, Guo RP, Jia WH, Yun JP, Chen MS, Zhang Y, Zhuang SM. A novel vascular pattern promotes metastasis of hepatocellular carcinoma in an epithelial-mesenchymal transition-independent manner. Hepatology 2015;62:452-65. [Crossref] [PubMed]
8. Renne SL, Woo HY, Allegra S, Rudini N, Yano H, Donadon M, Viganò L, Akiba J, Lee HS, Rhee H, Park YN, Roncalli M, Di Tommaso L. Vessels Encapsulating Tumor Clusters (VETC) Is a Powerful Predictor of Aggressive Hepatocellular Carcinoma. Hepatology 2020;71:183-95. [Crossref] [PubMed]
9. Tan PS, Nakagawa S, Goossens N, Venkatesh A, Huang T, Ward SC, Sun X, Song WM, Koh A, Canasto-Chibuque C, Deshmukh M, Nair V, Mahajan M, Zhang B, Fiel MI, Kobayashi M, Kumada H, Hoshida Y. Clinicopathological indices to predict hepatocellular carcinoma molecular classification. Liver Int 2016;36:108-18. [Crossref] [PubMed]
10. Lin WP, Xing KL, Fu JC, Ling YH, Li SH, Yu WS, Zhang YF, Zhong C, Wang JH, Chen ZY, Lu LH, Wei W, Guo RP. Development and Validation of a Model Including Distinct Vascular Patterns to Estimate Survival in Hepatocellular Carcinoma. JAMA Netw Open 2021;4:e2125055. [Crossref] [PubMed]
11. Feng Z, Li H, Zhao H, Jiang Y, Liu Q, Chen Q, Wang W, Rong P. Preoperative CT for Characterization of Aggressive Macrotrabecular-Massive Subtype and Vessels That Encapsulate Tumor Clusters Pattern in Hepatocellular Carcinoma. Radiology 2021;300:219-29. [Crossref] [PubMed]
12. Dennis C, Prince DS, Moayed-Alaei L, Remash D, Carr-Boyd E, Bowen DG, Strasser SI, Crawford M, Pulitano C, Kench J, McCaughan GW, McKenzie C, Liu K. Association between vessels that encapsulate tumour clusters vascular pattern and hepatocellular carcinoma recurrence following liver transplantation. Front Oncol 2022;12:997093. [Crossref] [PubMed]
13. Kawasaki J, Toshima T, Yoshizumi T, Itoh S, Mano Y, Wang H, Iseda N, Harada N, Oda Y, Mori M. Prognostic Impact of Vessels that Encapsulate Tumor Cluster (VETC) in Patients who Underwent Liver Transplantation for Hepatocellular Carcinoma. Ann Surg Oncol 2021;28:8186-95. [Crossref] [PubMed]
14. Fang JH, Xu L, Shang LR, Pan CZ, Ding J, Tang YQ, Liu H, Liu CX, Zheng JL, Zhang YJ, Zhou ZG, Xu J, Zheng L, Chen MS, Zhuang SM. Vessels That Encapsulate Tumor Clusters (VETC) Pattern Is a Predictor of Sorafenib Benefit in Patients with Hepatocellular Carcinoma. Hepatology 2019;70:824-39. [Crossref] [PubMed]
15. Wang JH, Li XS, Tang HS, Fang RY, Song JJ, Feng YL, Guan TP, Ruan Q, Wang J, Cui SZ. Vessels that encapsulate tumor clusters (VETC) pattern predicts the efficacy of adjuvant TACE in hepatocellular carcinoma. J Cancer Res Clin Oncol 2023;149:4163-72. [Crossref] [PubMed]
16. Xu W, Huang B, Zhang R, Zhong X, Zhou W, Zhuang S, Xie X, Fang J, Xu M. Diagnostic and Prognostic Ability of Contrast-Enhanced Unltrasound and Biomarkers in Hepatocellular Carcinoma Subtypes. Ultrasound Med Biol 2024;50:617-26. [Crossref] [PubMed]
17. Chen FM, Du M, Qi X, Bian L, Wu D, Zhang SL, Wang J, Zhou Y, Zhu X. Nomogram Estimating Vessels Encapsulating Tumor Clusters in Hepatocellular Carcinoma From Preoperative Gadoxetate Disodium-Enhanced MRI. J Magn Reson Imaging 2023;57:1893-905. [Crossref] [PubMed]
18. Zhu Y, Yang L, Wang M, Pan J, Zhao Y, Huang H, Sun K, Chen F. Preoperative MRI features to predict vessels that encapsulate tumor clusters and microvascular invasion in hepatocellular carcinoma. Eur J Radiol 2023;167:111089. [Crossref] [PubMed]
19. Fan Y, Yu Y, Hu M, Wang X, Du M, Guo L, Hu C. Imaging features based on Gd-EOB-DTPA-enhanced MRI for predicting vessels encapsulating tumor clusters (VETC) in patients with hepatocellular carcinoma. Br J Radiol 2021;94:20200950. [Crossref] [PubMed]
20. Itoh S, Yoshizumi T, Kitamura Y, Yugawa K, Iseda N, Shimagaki T, Nagao Y, Toshima T, Harada N, Kohashi K, Baba S, Ishigami K, Oda Y, Mori M. Impact of Metabolic Activity in Hepatocellular Carcinoma: Association With Immune Status and Vascular Formation. Hepatol Commun 2021;5:1278-89. [Crossref] [PubMed]
21. Lambin P, Leijenaar RTH, Deist TM, Peerlings J, de Jong EEC, van Timmeren J, Sanduleanu S, Larue RTHM, Even AJG, Jochems A, van Wijk Y, Woodruff H, van Soest J, Lustberg T, Roelofs E, van Elmpt W, Dekker A, Mottaghy FM, Wildberger JE, Walsh S. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol 2017;14:749-62. [Crossref] [PubMed]
22. Wu J, Liu A, Cui J, Chen A, Song Q, Xie L. Radiomics-based classification of hepatocellular carcinoma and hepatic haemangioma on precontrast magnetic resonance images. BMC Med Imaging 2019;19:23. [Crossref] [PubMed]
23. Shan QY, Hu HT, Feng ST, Peng ZP, Chen SL, Zhou Q, Li X, Xie XY, Lu MD, Wang W, Kuang M. CT-based peritumoral radiomics signatures to predict early recurrence in hepatocellular carcinoma after curative tumor resection or ablation. Cancer Imaging 2019;19:11. [Crossref] [PubMed]
24. Li Y, Zhang Y, Fang Q, Zhang X, Hou P, Wu H, Wang X. Radiomics analysis of [18F]FDG PET/CT for microvascular invasion and prognosis prediction in very-early- and early-stage hepatocellular carcinoma. Eur J Nucl Med Mol Imaging 2021;48:2599-614.
25. Liu K, Dennis C, Prince DS, Marsh-Wakefield F, Santhakumar C, Gamble JR, Strasser SI, McCaughan GW. Vessels that encapsulate tumour clusters vascular pattern in hepatocellular carcinoma. JHEP Rep 2023;5:100792. [Crossref] [PubMed]
26. Akiba J, Nakayama M, Sadashima E, Kusano H, Kondo R, Mihara Y, Naito Y, Mizuochi S, Yano Y, Kinjo Y, Tsutsui K, Kondo K, Sakai H, Hisaka T, Nakashima O, Yano H. Prognostic impact of vessels encapsulating tumor clusters and macrotrabecular patterns in hepatocellular carcinoma. Pathol Res Pract 2022;238:154084. [Crossref] [PubMed]
27. Dong X, Yang J, Zhang B, Li Y, Wang G, Chen J, Wei Y, Zhang H, Chen Q, Jin S, Wang L, He H, Gan M, Ji W. Deep Learning Radiomics Model of Dynamic Contrast-Enhanced MRI for Evaluating Vessels Encapsulating Tumor Clusters and Prognosis in Hepatocellular Carcinoma. J Magn Reson Imaging 2024;59:108-19. [Crossref] [PubMed]
28. Chen ZY, Guo ZX, Lu LH, Mei J, Lin WP, Li SH, Wei W, Guo RP. The predictive value of vessels encapsulating tumor clusters in treatment optimization for recurrent early-stage hepatocellular carcinoma. Cancer Med 2021;10:5466-74. [Crossref] [PubMed]
29. Yu Y, Fan Y, Wang X, Zhu M, Hu M, Shi C, Hu C. Gd-EOB-DTPA-enhanced MRI radiomics to predict vessels encapsulating tumor clusters (VETC) and patient prognosis in hepatocellular carcinoma. Eur Radiol 2022;32:959-70. [Crossref] [PubMed]
30. Santambrogio R, Opocher E, Costa M, Barabino M, Zuin M, Bertolini E, De Filippi F, Bruno S. Hepatic resection for "BCLC stage A" hepatocellular carcinoma. The prognostic role of alpha-fetoprotein. Ann Surg Oncol 2012;19:426-34. [Crossref] [PubMed]
31. Tyson GL, Duan Z, Kramer JR, Davila JA, Richardson PA, El-Serag HB. Level of α-fetoprotein predicts mortality among patients with hepatitis C-related hepatocellular carcinoma. Clin Gastroenterol Hepatol 2011;9:989-94. [Crossref] [PubMed]
32. Calderaro J, Couchy G, Imbeaud S, Amaddeo G, Letouzé E, Blanc JF, Laurent C, Hajji Y, Azoulay D, Bioulac-Sage P, Nault JC, Zucman-Rossi J. Histological subtypes of hepatocellular carcinoma are related to gene mutations and molecular tumour classification. J Hepatol 2017;67:727-38. [Crossref] [PubMed]
33. Kim KR, Moon HE, Kim KW. Hypoxia-induced angiogenesis in human hepatocellular carcinoma. J Mol Med (Berl) 2002;80:703-14. [Crossref] [PubMed]
34. Villa E, Critelli R, Lei B, Marzocchi G, Cammà C, Giannelli G, et al. Neoangiogenesis-related genes are hallmarks of fast-growing hepatocellular carcinomas and worst survival. Results from a prospective study. Gut 2016;65:861-9. [Crossref] [PubMed]
35. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, Bussink J, Monshouwer R, Haibe-Kains B, Rietveld D, Hoebers F, Rietbergen MM, Leemans CR, Dekker A, Quackenbush J, Gillies RJ, Lambin P. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 2014;5:4006. [Crossref] [PubMed]
36. Polverari G, Ceci F, Bertaglia V, Reale ML, Rampado O, Gallio E, Passera R, Liberini V, Scapoli P, Arena V, Racca M, Veltri A, Novello S, Deandreis D. (18)F-FDG Pet Parameters and Radiomics Features Analysis in Advanced Nsclc Treated with Immunotherapy as Predictors of Therapy Response and Survival. Cancers (Basel) 2020;12:1163. [Crossref] [PubMed]
37. Moazemi S, Erle A, Lütje S, Gaertner FC, Essler M, Bundschuh RA. Estimating the Potential of Radiomics Features and Radiomics Signature from Pretherapeutic PSMA-PET-CT Scans and Clinical Data for Prediction of Overall Survival When Treated with (177)Lu-PSMA. Diagnostics (Basel) 2021;11:186. [Crossref] [PubMed]
38. Wang G, Si Y, Liu J, Wang W, Yang J. Prognostic Value of Metabolic Parameters and Textural Features in Pretreatment (18)F-FDG PET/CT of Primary Lesions for Pediatric Patients with Neuroblastoma. Acad Radiol 2024;31:1091-101. [Crossref] [PubMed]
39. Yang J, Dong X, Wang F, Jin S, Zhang B, Zhang H, Pan W, Gan M, Duan S, Zhang L, Hu H, Ji W. A deep learning model based on MRI for prediction of vessels encapsulating tumour clusters and prognosis in hepatocellular carcinoma. Abdom Radiol (NY) 2024;49:1074-83. [Crossref] [PubMed]
40. Qu Q, Liu Z, Lu M, Xu L, Zhang J, Liu M, Jiang J, Gu C, Ma Q, Huang A, Zhang X, Zhang T. Preoperative Gadoxetic Acid-Enhanced MRI Features for Evaluation of Vessels Encapsulating Tumor Clusters and Microvascular Invasion in Hepatocellular Carcinoma: Creating Nomograms for Risk Assessment. J Magn Reson Imaging 2024;60:1094-110. [Crossref] [PubMed]
41. Meng XP, Tang TY, Zhou Y, Xia C, Xia T, Shi Y, Long X, Liang Y, Xiao W, Wang YC, Fang X, Ju S. Predicting post-resection recurrence by integrating imaging-based surrogates of distinct vascular patterns of hepatocellular carcinoma. JHEP Rep 2023;5:100806. [Crossref] [PubMed]
42. Wang YY, Dong K, Wang K, Sun Y, Xing BC. Effect of vessels that encapsulate tumor clusters (VETC) on the prognosis of different stages of hepatocellular carcinoma after hepatectomy. Dig Liver Dis 2023;55:1288-94. [Crossref] [PubMed]
43. Wang F, Wang Y, Qi L, Liang J, Xiao BH, Zhang C, Wáng YXJ, Ye Z. High correlation between Ki-67 expression and a novel perfusion MRI biomarker diffusion-derived vessel density (DDVD) in endometrial carcinoma. Magn Reson Imaging 2025;117:110324. [Crossref] [PubMed]