Prediction of the early hepatocellular carcinoma development in patients with chronic hepatitis B virus infection using gadoxetic acid-enhanced magnetic resonance imaging Authors: Mimi Tang, Danyang Xu, Huilin Jin, Chenyu Song, Xiaoqi Zhou, Huasong Cai, Lujie Li, Meicheng Chen, Yuxin Wu, Yanji Luo, Yuying Chen, Shi-Ting Feng Published: 2024 ABSTRACT AbstractBackgroundNon-hypervascular hypointense nodules (NHHNs) can transform into hypervascular hepatocellular carcinoma (HCC) during the long-term follow-up. However, the risk factors for NHHN hypervascular transformation in chronic hepatitis B virus (HBV)-infected populations are unknown. This study assessed the predictive value of gadoxetic acid-enhanced magnetic resonance imaging (MRI) for HCC development in patients with chronic HBV infection.MethodsA total of 86 patients with HBV infection who underwent gadoxetic acid-enhanced MRI at the First Affiliated Hospital of Sun Yat-sen University between January 2011 and July 2019 and were followed up for 2 years were retrospectively reviewed. Imaging features, including cirrhosis, steatosis, and NHHNs, were collected. Radiomics features were extracted from the entire liver. The HCC development predictive models were built based on each patient’s clinical data, MRI features, and radiomic features. We then collected the qualitative and quantitative features of each NHHN and investigated the risk factors of hypervascular transformation.ResultsThirteen patients developed HCC within two years. The risk factors for HCC development in patients with chronic HBV infection included older age, cirrhosis, and NHHNs. The MRI, radiomics, and integrated models developed all had an area under the curve (AUC) above 0.8. The potential risk factors for hypervascular transformation of NHHNs were the diameter of the NHHN (OR = 1.69, 95% CI:1.23, 2.32, P = 0.001) and the signal intensity (SI) ratio of the NHHN to the liver in the hepatobiliary phase (HBP SI ratio*10, OR = 0.36, 95% CI:0.11, 0.85, P = 0.044). The AUC of the hypervascular transformation model was 0.846 (95% CI:0.719, 0.972).ConclusionIn chronic HBV infection population, patients with older age, cirrhosis and NHHNs are more likely to develop HCC within two years. Models based on these factors or radiomic features can effectively predict HCC development. The diameter of the NHHNs and the signal intensity ratio of NHHN to the liver in the hepatobiliary phase are potential risk factors for the hypervascular transformation of NHHNs.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12885-024-13185-7.Keywords: Non-hypervascular hypointense nodules, Hepatocellular carcinoma, Radiomics, Magnetic resonance imaging Introduction Hepatocellular carcinoma (HCC) is a leading cause of cancer-related death worldwide. Hepatitis B virus (HBV) infection is the most prominent risk factor for the development of HCC, accounting for approximately half of the cases [1]. About 8–20% of patients with chronic HBV infection develop cirrhosis within five years of diagnosis [2]. The 5-year incidence of HCC in patients with cirrhosis is approximately 10% [3, 4]. Patients with early-stage HCC have a 5-year survival rate exceeding 70%, whereas the median survival of patients with advanced HCC is approximately one year [1, 5]. The key to early diagnosis and prognosis improvement is identifying patients with high levels of HCC risk and individualising HCC surveillance for such patients. Over the past few decades, several HCC risk scores based on risk factors such as cirrhosis, age, and male sex have been developed, including the Individual Prediction Model (IPM), Guide with Age, Gender, HBV DNA, Core Promoter Mutations, and Cirrhosis-HCC (GAG-HCC), and Page-B score [6]. However, the positive predictive values of most of these models are extremely low. Moreover, only a few can predict the location of HCC [7–9]. Hepatocarcinogenesis is a multistep process that culminates in the emergence of HCC [10]. Hepatic nodules that precede the emergence of HCC are classified as large regenerative nodules, low-grade dysplastic nodules (LGDN), and high-grade dysplastic nodules (HGDN) [11]. During multistep hepatocarcinogenesis, the arterial supply decreases at the LGDN stage to the early HCC stage, and the expression of organic anionic transporting polypeptides (OATP) diminishes progressively [12, 13]. These hypointense hepatocellular nodules detected in the hepatobiliary phase (HBP) of gadoxetic acid-enhanced magnetic resonance imaging (MRI) are depicted as non-hypervascular hypointense nodules (NHHNs) [14]. Recent studies have shown that NHHNs have a significant potential to transform into hypervascular HCC during long-term follow-up [15, 16]. However, whether NHHNs are risk factors for HCC development in chronic HBV-infected populations remains unknown. The risk factors for NHHN hypervascular transformation in chronic HBV-infected populations have not been reported in the literature. Many studies have reported that radiomics can be used to predict tumour behaviour and environment [17]. High-throughput radiomics may help identify risk factors for HCC development. Therefore, this study aimed to explore HCC risk factors in chronical HBV-infected populations using gadoxetic acid-enhanced MRI and the risk factors for NHHN hypervascular transformation into HCC in the chronically HBV-infected population. Materials and methods Population This study was approved by the Human Research Committee of the First Affiliated Hospital, Sun Yat-sen University (FAHSYSU) (No. [2017]258) and was conducted in accordance with the Declaration of Helsinki. The requirement for informed consent was waived due to the retrospective nature of the study. This retrospective study screened 4155 patients who underwent gadoxetic acid-enhanced MRI between January 2011 and July 2019 (Supplementary Fig. 1) at FAHSYSU. The inclusion criteria were patients with (1) chronic HBV infection and (2) at least two years of follow-up or patients who developed HCC within two years. The exclusion criteria were patients with (1) HCC or a history of HCC, (2) other malignancies or a history of malignancies, (3) who have had HCC within six months, and (4) severe artifacts influencing the quality of imaging. The patients were randomly allocated to the training and validation cohorts at a ratio of 7:3. MRI examination MRI was performed using a Magnetom Trio A Tim 3.0 T system (Siemens Healthcare Sector, Erlangen, Germany). The scanning range was from the top to the lower edge of the liver using an 8-channel phased-array coil as the receiver coil. The gadoxetic acid-enhanced MR scanning protocol included the following sequences: conventional T2-weighted image (T2WI) and T1-weighted image (T1WI) and fat-suppressed T1WI and T2WI. The dynamic enhanced images included three arterial phases, three portal venous phases, one delayed phase, and a 20-min HBP scanned in fat-suppressed VIBE T1WI. Gadoxetic acid (Primovist; Bayer Schering Pharma, Berlin, Germany) was injected into the cubital vein at a flow rate of 1 mL/s and dose of 0.025 mmol/kg, followed by flushing with 20 mL of normal saline. The MRI settings used in this study are detailed in Supplementary Table 1. MRI evaluation The qualitative and quantitative imaging features were independently evaluated by two abdominal radiologists with 7 and 10 years of experience each in MRI diagnosis. Both radiologists were blinded to the follow-up results. When the qualitative features evaluated by the radiologists were inconsistent, a consensus was reached after re-evaluation and discussion. For quantitative features, the average value measured by the two physicians was considered the result. The two radiologists independently evaluated the following qualitative features for each patient with chronic HBV infection (Fig. 1): (1) cirrhosis, defined as surface nodularity, global atrophy, segmental volume redistribution, regional or focal parenchymal contraction, parenchymal nodules, fibrotic scars, etc.; (2) steatosis, defined as the diffuse reduction of signals in the liver parenchyma in the opposed phase compared with the in phase sequence [18, 19]; (3) NHHNs, defined as solid nodules with hypointensity compared to the surrounding liver parenchyma in the HBP and hypo- or iso-enhanced nodules compared to the surrounding liver parenchyma during the dynamic arterial phase. Considering that the slice thickness of the post-contrast T1WI was 2 mm, nodules < 2 mm in diameter were excluded. Haemangiomas or hepatic cysts determined based on their classic enhancement pattern or non-enhancement on dynamic-phase images and hyperintensity on T2WI were excluded. MRI features. Cirrhosis in hepatobiliary phase (a); (b) NHHN (arrow) in hepatobiliary phase; steatosis, the diffuse reduction of signals in the liver parenchyma in the opposed phase (c) compared with the in phase sequence (d); NHHN (arrow) in the hepatobiliary phase (e) and washout in the portal phase (f). Abbreviations: NHHN, non-hypervascular hypointense nodule. Abbreviations: ROC, receiver operating characteristics curve The following qualitative feature were evaluated for each NHHN (Fig. 1): (1) washout in the portal phase. The following quantitative features were evaluated for each NHHN: (1) maximum nodule diameter, (2) mean value of the signal intensity (SI) of the nodule in the HBP, (3) SI of the nodule on unenhanced T1WI, and (4) SI ratio of the nodule to the liver in the HBP. The lesion should be measured at the level of the largest diameter where it is most clearly visible, and the main blood vessels should be avoided when measuring liver. For patients with multiple NHHNs, the characteristics of each NHHN mentioned above were individually recorded. In this study, the Liver Imaging Reporting and Data System was used to diagnose HCC during follow-up. Lesions classified as LR-5 were diagnosed as HCC. The specific diagnostic criteria were as follows: nodules > 20 mm with non-rim arterial-phase hyperenhancement and at least one of the following three signs: non-peripheral washout, enhancing capsule, or threshold growth; nodules > 10 mm with non-rim arterial phase hyperenhancement and one of the two signs: non-peripheral washout or threshold growth. Radiomics feature extraction Semiautomatic segmentation of the whole liver was performed in the HBP using VitreaCore (version 6.9.87.1) software. Radiomic features of the liver were extracted using Pyradiomics (https://github.com/AIM-Harvard/pyradiomics). The parameters for image normalisation were resampled pixel spacing (1, 1, 1) and bin width: 25. A total of 1688 imaging features were extracted, including six types of features: first-order statistics, Gray Level Co-occurrence Matrix (Glcm), Gray Level Run Length Matrix (Glrlm), Gray Level Size Zone matrix (Glszm), Gray Level Dependence Matrix (Gldm), and Neighbouring Gray Tone Difference Matrix (Ngtdm). Feature selection and model building To reduce redundancy, we selected the most useful radiomic features for HCC development using Selectkbest and least absolute shrinkage and selection operator (LASSO) regression. Logistic regression analysis was used to build the predictive radiomics model for HCC development based on the selected radiomic features. Statistical analysis The independent sample t-test or Mann–Whitney U test was used to compare clinical characteristics and MRI features between the groups in the training and validation cohorts for continuous variables, whereas the chi-square test or Fisher’s exact test was used for categorical variables. First, an HCC development predictive model was built based on each patient’s MRI features and clinical data. Univariate and multivariate logistic regression models were used to estimate the effects of various variables on the risk of HCC. Variables were selected into the logistic regression model based on forward selection, and those with P-values < 0.1 were selected into the training cohort and then validated in the validation cohort. We then attempted to integrate the radiomic features with the MRI features and clinical data to provide a combined clinical-MRI-radiomics model using logistic regression. Nomograms were constructed based on corresponding logistic models. The diagnostic ability of the model was assessed using the area under the receiver-operating characteristic curves (AUC). Accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. The receiver-operating characteristic curves of different models were compared using DeLong’s test. The goodness of fit of the predictive model was evaluated by Hosmer-Lemeshow goodness-of-fit test. Spearman’s correlation analysis was used to evaluate the associations between the MRI features, clinical data, and radiomic features. The risk factors for NHHN hypervascular transformation were analysed based on the qualitative and quantitative features of each nodule. Univariate and multivariate logistic regression models were used to estimate the effects of various variables on the risk. Considering that this part was exploratory and had a relatively small sample size, variables with P-values above 0.1 might be included in the multivariate analysis. Statistical significance was set at P < 0.05. Statistical analyses were performed using R software version 3.2.3 (Bell Laboratories; https://cran.r-project.org/bin/windows/base/old/ 3.2.3). Results Baseline characteristics Eighty-six patients were enrolled, with 60 and 26 patients in the training and validation cohorts, respectively. The median follow-up was 33.0 (8.0–73.7) months. HCC developed in 13 patients (15.1%) within two years of follow-up. Among these, four were confirmed by biopsy or surgical specimen pathology. The clinical and MRI features of the training and validation cohorts are presented in Table 1. The baseline characteristics of patients in the two cohorts were not significantly different except for cirrhosis. Regarding clinical and MRI features, patients with HCC were older and had cirrhosis and NHHNs (all P < 0.05; Supplementary Table 2). Among the HBV infected patients with NHHNs, 42.3% developed HCC within 2 years. Table 1Baseline characteristics between patients in the training and validation cohortsVariablesTotal(n = 86)Training cohort (n = 60)Validation cohort (n = 26)P-value1Sex0.456Male61 (70.9%)44 (73.3%)17 (65.4%)Female25 (29.1%)16 (26.7%)9 (34.6%)Age, years48 ± 1148 ± 1249 ± 90.541de novo HCC> 0.999Absent73 (84.9%)51 (85%)22 (84.6%)Present13 (15.1)9 (15)4 (15.4)Hepatic encephalopathy0.516Absent84 (97.7%)59 (98.3%)25 (96.2%)Present2 (2.3%)1 (1.7%)1 (3.8%)Steatosis0.205Absent74 (86%)54 (90%)20 (76.9%)Present12 (14%)6 (10%)6 (23.1%)NHHN0.943Absent60 (69.8%)42 (70%)18 (69.2%)Present26 (30.2%)18 (30%)8 (30.8%)Cirrhosis0.019Absent43 (50%)35 (58.3%)8 (30.8%)Present43 (50%)25 (41.7%)18 (69.2%)Abbreviations NHHN, non-hypervascular hypointense hepatic nodules1 T-test results for continuous variables with normal distribution, Mann–Whitney U test results for continuous variables with abnormal distribution, and Chi-square test or Fisher exact test results for categorical variablesContinuous variables with a normal distribution are presented as means (standard deviation, SD). Categorical variables are presented as N (%) according to the different levels Baseline characteristics between patients in the training and validation cohorts Abbreviations NHHN, non-hypervascular hypointense hepatic nodules 1 T-test results for continuous variables with normal distribution, Mann–Whitney U test results for continuous variables with abnormal distribution, and Chi-square test or Fisher exact test results for categorical variables Continuous variables with a normal distribution are presented as means (standard deviation, SD). Categorical variables are presented as N (%) according to the different levels Development and validation of HCC prediction model In the univariate and multivariate analyses of the clinical and MRI features, NHHN was a significant independent predictor of HCC development in the training cohort. Cirrhosis was a marginally significant predictor of the development of HCC (Table 2). According to Pearson’s test, age, cirrhosis, and NHHN were not correlated (Supplementary Fig. 2a). A nomogram was established based on the clinical and MRI features and their regression coefficients in the logistic model in the training cohort (Fig. 2a). The AUC of this MRI model was 0.886 (95% CI: 0.792–0.979) in the training cohort and 0.966 (95% CI: 0.890–1.000) in the validation cohort (Fig. 2b, c). The nomogram also showed good calibration (Hosmer-Lemeshow test P > 0.05). The accuracy, sensitivity, specificity, PPV, and NPV of the MRI model were 0.850, 0.889, 0.843, 0.500, and 0.977, respectively, in the training cohort and 0.885, 1.000, 0.864, 0.571, and 1.000 in the validation cohort. Table 2Univariate and multivariate analyses of clinical and MR imaging features in HCC development predictionUnivariate analysesMultivariate analysesVariablesOR95% CIP-valueOR95% CIP-valueSex (male)3.330.38–29.030.276Age1.071.01–1.140.0431.020.95–1.100.559Hepatic encephalopathy0.000.00 - Inf0.995Steatosis0.000.00 - Inf0.992Cirrhosis16.001.85–138.580.0128.920.91–87.290.060NHHN12.732.31–70.180.0037.381.19–45.770.032Abbreviations OR, odds ratio; CI, confidence interval; NHHN, non-hypervascular hypointense hepatic nodules Univariate and multivariate analyses of clinical and MR imaging features in HCC development prediction Abbreviations OR, odds ratio; CI, confidence interval; NHHN, non-hypervascular hypointense hepatic nodules The performance of the models for HCC development. The nomogram of the MRI model (a). The ROC curves of MRI model in the training (b) and validation cohort (c). The nomogram of the radiomics model (d). The ROC curves of radiomics model in the training (e) and validation cohort (f). The nomogram of the clinical-MRI-radiomics model (g). The ROC curves of clinical-MRI-radiomics model in the training (h) and validation cohort (i). Abbreviations: ROC, receiver operating characteristics curve A total of 1688 radiomics features of the whole liver were extracted (Supplementary Table 3), and 128 features were selected using Selectkbest. LASSO was used to reduce dependency further. Two features were selected to establish the radiomics model for HCC development prediction: auto_wavelet-HLL_gldm_HighGrayLevelEmphasis and auto_wavelet-LLH_glcm_ autocorrelation. The absolute value of the Pearson correlation coefficient between the features was < 0.5, indicating no strong correlation between these features (Supplementary Fig. 2b). A nomogram was established based on radiomic features and regression coefficients in the logistic model in the training cohort (Fig. 2d). The AUC of the radiomics model was 0.926 (95% CI: 0.854–0.998) in the training cohort and 0.830 (95% CI: 0.669–0.990) in the validation cohort (Fig. 2e, f). The nomogram also showed good calibration (Hosmer-Lemeshow test P > 0.05). The accuracy, sensitivity, specificity, PPV, and NPV of the radiomics model were 0.867, 0.889, 0.863, 0.533, and 0.978 in the training cohort and 0.769, 1.000, 0.727, 0.400, and 1.000, respectively, in the validation cohort. The radiomics model did not show improved performance compared to the MRI model in the training and validation cohorts (all P > 0.05). By integrating the selected radiomics features with all the clinical and MRI variables, we derived a combined clinical-MRI-radiomics model. In the univariate and multivariate analyses of the clinical, MRI and radiomics features, NHHN was still a significant independent predictor of HCC development in the training cohort (Supplementary Table 4). A nomogram was established based on the clinical, MRI, and radiomic features and their regression coefficients in the logistic model in the training cohort (Fig. 2g). The AUC of this integrated model was 0.972 (95% CI: 0.935–1.000) in the training cohort and 0.943 (95% CI: 0.850–1.000) in the validation cohort (Fig. 2h, i). The nomogram also showed good calibration (Hosmer-Lemeshow test P > 0.05). The accuracy, sensitivity, specificity, PPV, and NPV of the integrated model were 0.917, 1.000, 0.902, 0.643, and 1.000 in the training cohort and 0.885, 1.000, 0.864, 0.571, 1.000, respectively, in the validation cohort. The integrated model did not show improved performance compared to the MRI or radiomics models in the training and validation cohorts (all P > 0.05). Risk factors of NHHN hypervascular transformation Of the 86 patients, 26 (30.2%) had 57 NHHNs. Seventeen (29.8%) NHHNs in 11 patients progressed to HCC within two years. The mean diameter of all the nodules was 10.0 mm. This part was exploratory and had a relatively small sample size. Considering that some variables with P-values above 0.1 might be meaningful according to clinical experience and previous researches, like T1WI and washout [20, 21], we included these variables into multivariate analysis. In the univariate and multivariate analyses of the features, the maximum nodule diameter of the NHHN and the SI ratio of the nodule to the liver in the HBP might be the independent predictors of hypervascular transformation (Table 3). A nomogram was established based on these two independent predictors and their regression coefficients in the logistics model (Fig. 3a). The AUC of this hypervascular transformation model was 0.846 (95% CI: 0.719–0.972) (Fig. 3b). The nomogram also showed good calibration (Hosmer-Lemeshow test P > 0.05). The accuracy, sensitivity, specificity, PPV, and NPV of the hypervascular transformation model were 0.807, 0.824, 0.800, 0.636, and 0.914. Table 3Univariate and multivariate analyses of risk factors of NHHN hypervascular transformationUnivariate analysesMultivariate analysesVariablesOR95% CIP-valueOR95% CIP-valueNodule diameter1.461.17–1.83< 0.0011.691.23–2.320.001T1WI SI mean0.990.98–1.010.2290.980.96–1.000.063HBP SI mean1.001.00–1.010.6751.011.00–1.020.099HBP SI ratio0.040.00–3.940.1720.000.00–0.760.044HBP SI ratio*100.730.46–1.150.1720.360.11–0.850.044Washout in portal phase2.170.60–7.850.2390.610.09–4.390.625Abbreviations OR, odds ratio; CI, confidence interval; NHHN, non-hypervascular hypointense hepatic nodules. SD, standard deviation. SI, signal intensity. HBP, hepatobiliary phase Univariate and multivariate analyses of risk factors of NHHN hypervascular transformation Abbreviations OR, odds ratio; CI, confidence interval; NHHN, non-hypervascular hypointense hepatic nodules. SD, standard deviation. SI, signal intensity. HBP, hepatobiliary phase The performance of NHHN hypervascular transformation model. The nomogram of the NHHN hypervascular transformation model (a). The ROC curves of model (b). The ROC curves of diameter model (c). Abbreviations: ROC, receiver operating characteristics curve Using only the diameter to build the hypervascular transformation model, the AUC was 0.828 (95% CI: 0.686–0.970) (Fig. 3c). The Youden index–based cutoff value was 13.5 mm. The accuracy, sensitivity, specificity, PPV, and NPV of the hypervascular transformation model were 0.877, 0.588, 1.000, 1.000, and 0.851. Considering the need for surveillance, when the cutoff value was 9 mm, the accuracy, sensitivity, specificity, PPV, and NPV of the hypervascular transformation model were 0.754, 0.882, 0.700, 0.556, and 0.933. Figure 4 and Supplementary Fig. 3 showed the cases with and without hypervascular transformation, respectively. An NHHN with hypervascular transformation. The NHHN was showed as isointensity on T1WI (a), without hyperenhancement in arterial phase (b) and washout in portal phase (c), only detected in hepatobiliary phase (d). After one year’s follow up, the NHHN was showed as hypointensity on T1WI (e), hyperenhancement in arterial phase (f), washout in portal phase (g) and hypointensity in hepatobiliary phase (h). The diameter of the NHHN was 14 mm. The signal intensity ratio of the nodule to the liver in the hepatobiliary phase was 0.583. The predicted value originated from the hypervascular transformation model was 0.639, higher than the cutoff value 0.287 Discussion This study explored risk factors for HCC in chronically HBV-infected population using gadoxetic acid-enhanced MRI, including age, cirrhosis, and NHHNs. Subsequently, the MRI, radiomics, and combined models were established. All these models had excellent discrimination, and all AUCs were above 0.8. All models had sensitivities close to or up to 1 in the training and validation cohorts, which is necessary for surveillance. Our findings also showed that the maximum nodule diameter of the NHHN and the SI ratio of the nodule to the liver in the HBP were potential predictors of hypervascular transformation into HCC in the chronically HBV-infected population. To our knowledge, this is the first study to assess HCC risk using gadoxetic acid-enhanced MRI in a population with chronic HBV infection. Several studies have reported on the development of HCC risk scores. In Fan et al.‘s study, the 3-year and 5-year cumulative incidences of HCC were < 20%, even in the high-risk group [7]. The PPV was set to < 20%. In Yang et al.‘s study, the 3-year incidence of HCC was 23.6% in the high-risk group [22]. In the present study, 42.3% of the HBV infected patients with NHHNs developed HCC within two years. Our models achieved a higher PPV while maintaining very high sensitivity. The PPVs of all models were approximately 50%. Previous studies have shown that patients with cirrhosis are more likely to develop HCC during follow-ups. The incidence rates of HCC are 0.6 per 100 person-years for those with chronic HBV infection without cirrhosis and 3.7 per 100 person-years for those with compensated cirrhosis [23]. Our study also showed that cirrhosis was a risk factor for HCC development. HCC occurred in 27.9% of patients with cirrhosis in our study. Although HCC can occur in the absence of cirrhosis in people infected with HBV, 70–90% of patients with HCC usually have cirrhosis [24]. In patients with chronic HBV infection, hepatitis B does not directly kill cells. Recognition of the virus as a foreign antigen activates host immunity to target and destroy the infected liver cells. Repeated inflammation and liver injury can lead to liver fibrosis and HCC [25]. Liver biopsy remains the gold standard to determine liver inflammation and fibrosis. However, biopsy is associated with significant risks, including bleeding (0.60%) and damage to other organs (0.08%), and sampling errors may underestimate the presence of the disease. In this study, cirrhosis was evaluated using imaging examination [26, 27]. Evaluation of cirrhosis is more comprehensive and accurate when based on liver morphology, signals, and manifestations of decompensation. Our study also showed that NHHNs are independent risk factors for HCC development. NHHNs reflect a decrease in arterial blood supply and OATP expression during hepatocarcinogenesis. In Joo et al.‘s study, NHHNs mainly corresponded to progressive HCCs, early HCCs, and HGDN [28]. Because there was only one arterial phase in this study, the sensitivity of HCC diagnosis was inevitably reduced [29]. Nevertheless, this study shows that the essence of NHHNs may be mainly precancerous lesions of HCC, such as early HCC, HGDN, and LGDN. The annual incidence of HCC progression is approximately 10% in LGDN and 20% in HGDN [30]. Considering the high potential for hypervascular transformation, intensive follow-up every 3–6 months is recommended according to the Japanese clinical practice guidelines for HCC [31]. We also investigated the risk factors for NHHN hypervascular transformation in a chronically HBV-infected population. Previous studies have reported that the size of the NHHN is associated with hypervascularisation [14]. Takashi et al. showed that the proportion of hypervascular transformation in NHHNs with a diameter > 15 mm was significantly higher than that in NHHNs with a diameter < 15 mm, similar to the present study’s results [14]. The International Consensus Group for Hepatocellular Neoplasia also points out that the development of HCC is gradual, and one of its important features is an increase in nodule diameter [11]. During this process, the expression of OATP8, which is responsible for the transport of contrast agents on the surface of hepatocytes, also decreases gradually with an increase in hepatocyte dysplasia, resulting in a gradual decrease in NHHN signals [32]. Wang et al. reported significant differences in the SI ratios of lesions-to-parenchyma in HBP images between the hypervascularised and maintained non-hypervascular groups at the endpoint of a three-time follow-up [33]. In our study, the radiomics model was based on the images in the HBP from gadoxetic acid-enhanced MRI. Unlike traditional Gd-DTPA-enhanced MRI, after injection, approximately 50% of the disodium gadolinate is transported to the liver cells through the OATP8 on the liver cell membrane. Hepatocyte absorption peaked 20 min after injection, producing a hepatobiliary-specific phase. The signal difference between the liver and surrounding tissue in HBP images is significantly different, which is more favourable for whole-liver segmentation. The morphology and function of the liver in patients with HBV infection are altered, and regenerated and dysplastic nodules are observed in cirrhosis. NHHNs can be highlighted in the HBP, which is conducive to comprehensive analysis and feature extraction of the liver parenchyma. Our results suggest that Glcm and Gldm are associated with the development of HCC in patients with HBV infection. Glcm can usually distinguish the level difference of an image from the complexity, direction, and similarity of the texture. This may indicate that the liver parenchyma of these patients has greater hierarchical differences and more complex textures. Gldm is based on the grayscale relationship between the central pixel or voxel and its neighbourhood. This feature measures the distribution of higher gray values, indicating the concentration of higher gray values in the image. The combined clinical-MRI-radiomics model did not show statistically significant improved diagnostic performance than the MRI model in predicting HCC development in patients with chronic HBV infection, which indicated the application of radiomics may be unnecessary. This raises concerns about the clinical utility of incorporating more complex radiomics features when simpler MRI metrics already provide robust predictions. This study had some limitations. First, as a retrospective study, some patients with chronic HBV infection were not included because they did not undergo gadoxetic acid-enhanced MRI, which may have led to potential selection bias. In our research, with a subset of patients being outpatients, the data we were able to collect was quite limited. For example, there were more than one-third patients lacked of AFP test. We could not include tumour markers such as AFP into our analysis. The second was the small sample size, especially with the de novo HCC group, and lack of external validation due to limited clinical resources and the nature of this study, which may lead to unsatisfactory robustness of some results. Therefore, it is necessary to collect cases in the future and expand the sample size to validate the results. Third, we built this HCC development prediction model based on gadoxetic acid-enhanced MRI. In current guidelines, ultrasound is recommended for screening [1]. Since the cost of MRI is likely to decrease in the future and people’s health awareness is increasing, Gadoxetic acid-enhanced MRI is getting more and more widely used in screening. Conclusion In chronic HBV infection population, patients with older age, cirrhosis, and NHHNs are more likely to develop HCC within two years. Models based on these factors or radiomic features can effectively predict HCC development. The size of the NHHNs and the SI ratio of the NHHN to the liver in the HBP are potential independent risk factors for the hypervascular transformation of NHHNs. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1: Supplementary Fig. 1. Flow diagram of the study population. Abbreviations: MRI, magnetic resonance imaging; HCC, hepatocellular carcinoma. Supplementary Material 2: Supplementary Fig. 2. Pearson correlation coefficient between the clinical, MRI features (a) and radiomics features (b). Abbreviations: NHHN, non-hypervascular hypointense nodule. Supplementary Material 3: Supplementary Fig. 3. An NHHN without hypervascular transformation. The NHHN was showed as isointensity on T1WI (a), without hyperenhancement in arterial phase (b) and washout in portal phase (c), only detected in hepatobiliary phase (d). After more than two years’ follow up, the NHHN was showed as isointensity on T1WI (e), without hyperenhancement in arterial phase (f) and washout in portal phase (g), only detected in hepatobiliary phase (h). The diameter of the NHHN was 6 mm. The signal intensity ratio of the nodule to the liver in the hepatobiliary phase was 0.373. The predicted value originated from the hypervascular transformation model was 0.156, lower than the cutoff value 0.287. Supplementary Material 4 Acknowledgements Not applicable. Abbreviations Area Under the Curve Hepatobiliary Phase Hepatitis B Virus Hepatocellular Carcinoma Magnetic Resonance Imaging Non-Hypervascular Hypointense Nodule Negative Predictive Value Organic Anionic Transporting Polypeptides Positive Predictive Value Signal Intensity T1-Weighted Image T2-Weighted Image Author contributions S.T.F., Y.C. and Y.L. conceived and designed the study. M.T. and Y.L. performed the MRI analyses. D.X. and C.S. undertook the radiomics model building. M.T., X.Z., H.C., L.L., M.C. and Y.W. collected clinical information. M.T. and H.J. undertook the statistical analyses. M.T., H.J. and Y.C. wrote the first draft of the manuscript. S.T.F., Y.C. and Y.L. reviewed the paper, and made critical revision of the manuscript for important intellectual content. All authors confirm that they had full access to all the data in the study. All authors read and approved the final version of the manuscript. Funding The work of the authors is supported by National Natural Science Foundation of China (No. 82471948, 82271958), Natural Science Foundation of Guangdong Province (2024A1515011968, 2024A1515012149, 2023A1515011097). Data availability The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate This study was approved by the Human Research Committee of the First Affiliated Hospital, Sun Yat-sen University (FAHSYSU) (No. [2017]258) and was conducted in accordance with the Declaration of Helsinki. The requirement for informed consent was waived due to the retrospective nature of the study. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Mimi Tang, Danyang Xu and Huilin Jin contributed equally to this work and share the first authorship. Contributor Information Yanji Luo, Email: luoyj26@mail.sysu.edu.cn. Yuying Chen, Email: chenyy593@mail.sysu.edu.cn. Shi-Ting Feng, Email: fengsht@mail.sysu.edu.cn. References Associated Data Supplementary Materials Supplementary Material 1: Supplementary Fig. 1. Flow diagram of the study population. Abbreviations: MRI, magnetic resonance imaging; HCC, hepatocellular carcinoma. Supplementary Material 2: Supplementary Fig. 2. Pearson correlation coefficient between the clinical, MRI features (a) and radiomics features (b). Abbreviations: NHHN, non-hypervascular hypointense nodule. Supplementary Material 3: Supplementary Fig. 3. An NHHN without hypervascular transformation. The NHHN was showed as isointensity on T1WI (a), without hyperenhancement in arterial phase (b) and washout in portal phase (c), only detected in hepatobiliary phase (d). After more than two years’ follow up, the NHHN was showed as isointensity on T1WI (e), without hyperenhancement in arterial phase (f) and washout in portal phase (g), only detected in hepatobiliary phase (h). The diameter of the NHHN was 6 mm. The signal intensity ratio of the nodule to the liver in the hepatobiliary phase was 0.373. The predicted value originated from the hypervascular transformation model was 0.156, lower than the cutoff value 0.287. Supplementary Material 4 Data Availability Statement The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.