Abstract
Objectives
This study aims to develop and validate a predictive nomogram for early recurrence in hepatocellular carcinoma (HCC), utilizing gadoxetic acid-enhanced MRI and intravoxel incoherent motion (IVIM) imaging to improve preoperative assessment and decision-making.
Materials and methods
From March 2018 and June 2022, a total of 245 patients with pathologically confirmed HCC, who underwent preoperative gadoxetic acid-enhanced MRI and IVIM, were retrospectively enrolled from two hospitals. These patients were divided into a training cohort (n = 160) and a validation cohort (n = 85). All patients were followed until death or the last follow-up date, with a minimum follow-up period of two years. Clinical indicators and pathologic information were compared between train cohort and validation cohort. Radiological features and diffusion parameters were compared between recurrence and non-recurrence groups using the chi-square test, Mann-Whitney U test and independent sample t test in training cohort. Univariate and multivariate analyses were performed to identify significant clinical-radiological variables associated with early recurrence in the training cohort. Based on these findings, a predictive nomogram integrating risk factors and diffusion parameters was developed. The predictive performance of the nomogram was evaluated in both the training and validation cohorts.
Results
No statistically significant difference in clinical and pathologic characteristics were observed between the training and validation cohorts. In training cohort, significant differences were identified between the recurrence and non-recurrence groups in tumor size, nodule-in-nodule architecture, mosaic architecture, non-smooth tumor margin, intratumor necrosis, satellite nodule, and peritumoral hypo-intensity in the hepatobiliary phase (HBP). The results of multivariate analysis identified tumor size (HR, 1.435; 95 % CI, 0.702–2.026; p < 0.05), mosaic architecture (HR, 0.790; 95 % CI, 0.421–1.480; p < 0.05), non-smooth tumor margin (HR, 1.775; 95 % CI, 0.941–3.273; p < 0.05), intratumor necrosis (HR, 1.414; 95 % CI, 0.807–2.476; p < 0.05), satellite nodule (HR, 0.648; 95 % CI, 0.352–1.191; p < 0.01), peritumoral hypo-intensity on HBP (HR, 2.786; 95 % CI, 1.141–6.802; p < 0.001) and D (HR, 0.658; 95 % CI,0.487–0.889; p < 0.01) were the independent risk factor for recurrence. The nomogram exhibited excellent predictive performance with C-index of 0.913 and 0.875 in the training cohort and validation cohort, respectively. Also, based on the nomogram score, the patients were classified according to risk factor and the Kaplan-Meier curve analysis also showed that the nomogram had a good predictive efficacy.
Conclusion
The nomogram, integrating radiological risk factors and diffusion parameters, offers a reliable tool for preoperative prediction of early recurrence in HCC patients.
Key Words: Hepatocellular carcinoma, Recurrence, Magnetic resonance imaging, Diffusion
1. Introduction
Hepatocellular carcinoma (HCC) accounts for 80–90 % of primary liver cancers and is the fourth leading cause of cancer-related mortality worldwide [1]. Hepatectomy remains the primary curative treatment for early-stage HCC in patients with sufficient liver functional [2]. Despite advancements in surgical techniques and perioperative management, the survival rate of HCC patients has improved, yet long-term outcomes continue to be impacted by a high recurrence rate [3], [4]. According to the latest guidelines for liver cancer diagnosis and treatment, recurrence within two years is defined as early recurrence, while recurrence after two years is considered late recurrence. Importantly, early recurrence is associated with significantly worse prognosis compared to late recurrence [5]. Several studies have identified potential risk factors for early recurrence, including specific serologic markers, microvascular invasion (MVI), vessels encapsulating tumor clusters (VETC), and poor histologic differentiation [6], [7]. However, these risk factors can typically only be assessed postoperatively through pathological examination, limiting their utility in preoperative decision-making.
Recent studies suggest that the biological characteristics and heterogeneity of HCC can be effectively assessed through imaging, particularly gadoxetic acid-enhanced MRI [8], [9]. As a liver-specific contrast agent, gadoxetic acid (GA) can be taken up by hepatocytes, and GA-MRI not only captures hemodynamic features of HCC, but also provide insight into liver function on hepatobiliary phase (HBP), making its imaging features highly informative in predicting HCC heterogeneity [10]. For instance, factors such as larger tumor size, non-smooth tumor margins, and peritumoral hypo-intensity on hepatobiliary phase (HBP) have been associated with MVI, while the presence of heterogeneous enhancement with septations and intra-tumoral artery are significant predictor of VETC [11], [12], [13]. However, in the presence of severe liver dysfunction, diminished GA uptake leads inadequate enhancement on HBP, making the above features less likely to be demonstrated.
The traditional diffusion-weighted imaging (DWI) parameter, the apparent diffusion coefficient (ADC), quantifies the diffusion motion of water molecules and reflect tumor microstructure. However, the ADC calculated from a mono-exponential model which ignored the microcirculation perfusion effect in tissue. Intravoxel incoherent motion model (IVIM), proposed by Le Bihan et al., posits that the MR signal attenuation is influenced by both true molecular diffusion and perfusion from the blood microcirculation by using a bi-exponential model [14]. Previous study has reported that IVIM outperforms conventional DWI in accurately in predicting MVI and histological grade of HCC, however, further validation is required [15].
While these findings are promising, the interpretation of imaging features is often subject to observer bias, and the lack of external validation for quantitative results undermines the reliability of these conclusions [16]. Furthermore, limited studies have explored the combined use of gadoxetic acid-enhanced MRI and diffusion parameters in predicting early recurrence.
Thus, this dual-center study aimed to develop an easy-to-use nomogram that integrates radiological features and diffusion parameters to predict early recurrence of HCC and to evaluate its predictive efficacy through internal and external validation.
2. Materials and methods
2.1. Patient population
This two-center retrospective study was approved by the institutional review board at each participating center (approval number NSMC201801124 and CDRAD20191112, respectively), and the requirement for patient consent was waived. Consecutive patients with pathologically confirmed HCC who underwent gadoxetic acid-enhanced MRI and IVIM within one month prior to curative resection were reviewed. The patients were divided into two separate cohorts: the training cohort and the validation cohort. The training cohort consisted of HCC patients from Hospital A (Affiliated Hospital of North Sichuan Medical College, scanner: Discovery MR750) between March 2018 and June 2022. The validation cohort comprised HCC patients from Hospital B (The Second People’s Hospital of Chengdu, scanner: MAGNETOM Vida) between February 2020 and June 2022. Patients were excluded if they met any of the following criteria: (a) history of prior treatment, (b) tumors too small (<1 cm) to measure, (c) poor imaging quality due to motion and susceptibility artifacts that could compromise accurate measurement, or (d) lost to follow-up. Ultimately, 245 patients (160 in the training cohort and 85 in the validation cohort) were included in this study (Fig. 1).
Fig. 1.
Flow chart of the study population. HCC, hepatocellular carcinoma; RFA, radiofrequency ablation; TACE, transcatheter arterial chemoembolization.
Clinical and laboratory data, including sex, age, etiology of liver disease, presence or absence of cirrhosis, serum alpha-fetoprotein (AFP) levels, and Child-Pugh classification were extracted from medical records. Pathological characteristics, including Edmondson-Steiner grade, MVI and VETC were evaluated by consensus between 2 experienced pathologists.
2.2. MRI examination
All MRI examinations were performed using two 3.0 T systems (Discovery MR750, GE Healthcare, USA; MAGNETOM Vida, Siemens Healthineers, Germany) with a phase-array torso coil. The baseline MRI protocol included fast spin-echo T2-weighted imaging, in-phase and opposed-phase imaging dual-echo fast spoiled gradient-echo T1-weighted. Conventional DWI was acquired by using a single-shot echo-planner imaging (SE-EPI) sequence with three b values (0, 50 and 800 s/mm2). Multi-b-value IVIM was also conducted with eleven b values ranging from 0 to 1000 s/mm2 (0, 10, 20, 50, 100, 120, 150, 200, 400, 800 and 1000 s/mm2), with the number of excitations (NEX) for each b-value was 1, 4, 4, 2, 2, 1, 1, 1, 2, 4 and 6, respectively. Parallel imaging (array spatial sensitivity encoding technique, ASSET) was utilized for both conventional DWI and IVIM to minimize image distortion. For gadoxetic acid (Primovist, Bayer Pharma AG, Berlin, Germany)-enhanced MRI, the following images were obtained using a T1-weighted three-dimensional (3D) fast spoiled gradient-echo (liver acquisition with volume acceleration; GE Healthcare) sequence: unenhanced phase, enhanced arterial phase (AP, 20–40 s), portal venous phase (PVP, 50–60 s), delayed phase (DP, 3 min), and HBP (20 min). The contrast agent was administered intravenously at a total dose of 0.025 mmol/kg body weight with a rate of 1 ml/s, followed by a 20 ml saline flush. The detailed parameters of each acquisition sequence are shown in Supplementary Table 1.
2.3. Imaging analysis
All MRI images were reviewed independently by two radiologists (D.G. and L.P.L., with 7 and 11 years of experience in liver MRI, respectively), who were blinded to the clinical, pathological and follow up information. In case of any discrepancy, a consensus was reached through discussion.
The two radiologists independently evaluated the following imaging features for each HCC: (a) tumor burden, including tumor size and number; (b) features reported in the Liver Imaging Reporting and Data System (LI-RADS) v2018 associated with recurrence; and (c) non-LI-RADS features previously linked to prognosis, such as non-smooth tumor margin, intra-tumoral arteries, intra-tumoral necrosis, and peritumoral hypo-intensity on HBP [17], [18].
For qualitative assessment, first, DWI and IVIM post-processing were performed using Functool software (version AW 4.6, GE Healthcare) to generate parametric maps. Second, regions of interest (ROIs) were manually placed on these maps at the level of the maximum tumor diameter and traced along the tumor borders to encompass the largest tumor area. Necrotic or hemorrhagic areas, as well as artifacts, were avoided by referring to T2-weighted images and contrast-enhanced T1-weighted images. The parameters derived from DWI and IVIM model were calculated using the following equation:
| DWI equation: Sb/S0 = exp (-b × ADC) |
| IVIM equation: Sb/S0= (1-f) × exp (-b × D) + f × exp (-b × D*) |
where Sb and S0 represent the signal intensity with and without diffusion gradient, respectively. ADC is the apparent diffusion coefficient, D is the diffusion coefficient, representing pure molecular diffusivity, D* is the perfusion parameter, representing pseudo-diffusion coefficient, and f is the proportion of perfusion linked to microcirculation. D, D* and f were estimated using the Levenberg-Marquardt algorithm for IVIM fitting [19].
2.4. Follow-up plan
Postoperative follow-up included serum AFP, ultrasound, CT, or MRI within one month after surgery and every 3 months for the first two years, followed by semi-annual follow-ups thereafter. Patients were monitored until death or last follow-up date (June 30, 2024), with a minimum follow-up duration of two years.
Early recurrence was defined as intrahepatic and/or extrahepatic recurrence of HCC within two years after curative resection. Therefore, early recurrence-free survival (RFS) was defined as from the date of surgery to the date of recurrence or the last follow-up date without recurrence within two years.
2.5. Statistical analysis
Statistical analyses were performed using SPSS v. 23.0 software (Chicago, IL), and plots were generated using R package v.4.2.0. Continuous data were tested for normality using the Kolmogorov-Smirnov test and are presented as mean ± standard deviation (SD) or median ± interquartile range (IQR). Categorical data were analyzed using the chi-square test (or Fisher’s exact test when a subgroup contained fewer than 5 patients). Interobserver variability was assessed using Cohen’s kappa for categorical variables and intraclass correlation coefficient (ICC) with a two-way random effects model for continuous variables.
Diffusion parameters derived from DWI and IVIM were compared between the recurrence and non-recurrence groups. For normally distributed data, Student’s t-test was used, whereas the Mann-Whitney U test was applied to data that did not meet the assumptions of normality. A p-value of less than 0.05 was considered statistically significant.
Cox proportional hazards regression analysis was used to assess the imaging features associated with HCC recurrence. Univariate regression analysis was initially performed, followed by multivariate regression using a stepwise forward selection method to identify independent risk factors for recurrence. Variables with a p-value < 0.05 in univariate analysis were included in the multivariate model. Based on the results of the multivariate analysis, a nomogram incorporating clinical, radiological, and diffusion parameters was developed and validated to predict early recurrence. The predictive performance of the nomogram was evaluated using the concordance index (C-index) and calibration with 1000 bootstrap samples to mitigate overfitting bias. Recurrence rates were estimated using Kaplan-Meier curves, and differences in recurrence rates across model-predicted risk strata were assessed using the log-rank test.
3. Results
3.1. Clinicopathologic characteristics
There were no significant differences in clinicopathologic characteristics of patients between the training and validation cohorts, including patients age, gender, etiology, laboratory test results, pathologic grade and MVI and VETC states. The follow-up duration ranged from 24 to 72 months in the training cohort and from 24 to 48 months in the validation cohort. A detailed comparison of characteristics is summarized in Table 1.
Table 1.
Comparison of patient characteristics in training and validation cohort.
| Characteristics | Total (n = 245) |
Training cohort (n = 160) |
Validation cohort (n = 85) |
p value |
|---|---|---|---|---|
| Age (year) | 54.2 ± 7.5 | 54.4 ± 7.8 | 53.8 ± 7.0 | 0.593 |
| Gender | 0.498 | |||
| Male | 202 (82.4) | 130 (81.3) | 72 (84.7) | |
| Female | 43 (17.6) | 30 (18.7) | 13 (15.3) | |
| Etiology of liver disease | 0.919 | |||
| HBV | 233 (95.1) | 152 (95.0) | 81 (95.2) | |
| HCV | 4 (1.6) | 3 (1.9) | 1 (1.2) | |
| Alcohol | 3 (1.2) | 2 (1.2) | 1 (1.2) | |
| Other | 5 (2.1) | 3 (1.9) | 2 (2.4) | |
| Liver cirrhosis | 0.766 | |||
| Presence | 223 (91.0) | 145 (90.6) | 78 (91.7) | |
| Absence | 22 (9.0) | 15 (9.4) | 7 (8.3) | |
| Serum AFP level | 0.283 | |||
| ≤ 400 ng/ml | 168 (68.6) | 106 (66.3) | 62 (72.9) | |
| > 400 ng/ml | 75 (31.4) | 54 (33.7) | 23 (37.1) | |
| Child-Pugh score | 0.801 | |||
| A | 240 (98.0) | 157 (98.1) | 83 (97.6) | |
| B | 5 (2.0) | 3 (1.9) | 2 (2.4) | |
| Edmondson-Steiner grade | 0.374 | |||
| Grade 1 | 46 (18.8) | 27 (16.9) | 19 (22.4) | |
| Grade 2 | 92 (37.6) | 59 (36.9) | 33 (38.8) | |
| Grade 3 | 69 (28.1) | 49 (30.6) | 20 (23.5) | |
| Grade 4 | 38 (15.5) | 25 (15.6) | 13 (15.3) | |
| MVI | 0.325 | |||
| Presence | 102 (41.6) | 63 (39.4) | 39 (45.8) | |
| Absence | 143 (58.4) | 97 (60.6) | 46 (54.2) | |
| VETC | 0.098 | |||
| Presence | 52 (21.2) | 39 (24.4) | 13 (15.3) | |
| Absence | 193 (78.8) | 121 (75.6) | 72 (84.7) | |
| Early occurrence | 0.411 | |||
| Presence | 95 (38.8) | 64 (40.0) | 31 (36.5) | |
| Absence | 150 (61.2) | 96 (60.0) | 54 (63.5) | |
| Time to early recurrence (month) | 19.2 ± 3.8 | 19.4 ± 3.7 | 18.8 ± 4.2 | 0.476 |
Unless otherwise specified, data are number of patients with percentage in parentheses.
Categorical variables were compared by using the χ2 test or Fisher exact test.
HBV, Hepatitis B virus; HCV, Hepatitis C virus; AFP, alpha-fetoprotein; MVI, microvascular invasion; VETC, vessels that encapsulate tumor clusters
3.2. Interobserver agreement for radiologic features and parameters in two cohorts
Excellent interobserver agreement was observed between the two radiologists for imaging features, with all kappa values exceeding 0.8. Similarly, the interobserver agreement for tumor size (ICC, 0.978; 95 % CI, 0.963–0.989), ADC (ICC, 0.893; 95 % CI, 0.858–0.932), D (ICC, 0.913; 95 % CI, 0.901–0.928), D* (ICC, 0.877; 95 % CI, 0.846–0.903), and f (ICC,0.859 95 % CI, 0.834–0.882) was also excellent
3.3. Comparison of imaging features and diffusion parameters between recurrence and no recurrence in training cohort
There were statistically significant differences between the recurrence group and no recurrence group in tumor size, nodule-in-nodule architecture, mosaic architecture, non-smooth tumor margin, intratumor necrosis, satellite nodule, and peritumoral hypo-intensity on HBP (all p < 0.05; Fig. 2 and Supplementary Fig), whereas other imaging features, such as non-rim hyperenhancement on AP, non-peripheral washout, intratumor hemorrhage, etc., did not show statistically significant (all p > 0.05) (Table 2). The comparison results of the validation cohort are shown in Supplementary Table 2.
Fig. 2.
A 61-year-old male patient with a 5.2 cm HCC was treated with curative resection. Tumor recurrence occurred 5 months after surgery. (a) The lesion shows hyperintensity on T2 weighted imaging; (b) Heterogeneously rim enhancement on arterial phase; (c) The lesion demonstrates non-smooth tumor margin (white arrow), peritumoral hypo-intensity (black arrow) and satellite nodule (arrow head) on hepatobiliary phase; (d) D map, D value for the lesion was 0.727 × 10−3 mm2/s; (e) D* map, D* value for the lesion was 29.86 × 10−3 mm2/s; (f) f map, f value for the lesion was 25.23 %.
Table 2.
Comparison of imaging features according to early recurrence in training cohort.
| Imaging features | Total (n = 160) |
Early Recurrence (n = 64) |
No Early Recurrence (n = 96) |
p value |
|---|---|---|---|---|
| LI-RADS major features | ||||
| Tumor diameter (cm) | 3.97 ± 1.22 | 4.25 ± 1.36 | 3.60 ± 0.96 | < 0.001 |
| Non-rim hyperenhancement on AP | 0.674 | |||
| Presence | 151 (94.3) | 61 (95.3) | 90 (93.8) | |
| Absence | 9 (5.7) | 3 (4.7) | 6 (6.2) | |
| Non-peripheral washout | 0.658 | |||
| Presence | 148 (92.5) | 59 (92.2) | 89 (92.7) | |
| Absence | 12 (7.6) | 5 (7.8) | 7 (7.3) | |
| LI-RADS ancillary features | ||||
| Intratumor hemorrhage | 0.793 | |||
| Presence | 26 (16.3) | 11 (17.2) | 15 (15.6) | |
| Absence | 134 (83.7) | 53 (82.8) | 81 (84.4) | |
| Nodule-in-nodule architecture | 0.014 | |||
| Presence | 18 (11.3) | 12 (18.8) | 6 (6.3) | |
| Absence | 142 (88.7) | 52 (81.2) | 90 (93.7) | |
| Corona enhancement | 0.242 | |||
| Presence | 35 (21.9) | 17 (26.6) | 18 (18.8) | |
| Absence | 125 (78.1) | 47 (73.4) | 78 (81.2) | |
| Mosaic architecture | < 0.001 | |||
| Presence | 58 (36.3) | 36 (56.3) | 22 (22.9) | |
| Absence | 102 (63.7) | 28 (43.7) | 74 (77.1) | |
| No LI-RADS features | ||||
| Non-smooth tumor margin | < 0.001 | |||
| Presence | 85 (53.1) | 45 (70.3) | 40 (41.7) | |
| Absence | 75 (46.9) | 19 (29.7) | 56 (58.3) | |
| Intratumor necrosis | 0.002 | |||
| Presence | 62 (38.8) | 34 (53.1) | 28 (29.2) | |
| Absence | 98 (61.2) | 30 (46.9) | 68 (70.8) | |
| Satellite nodule | < 0.001 | |||
| Presence | 36 (22.5) | 27 (42.2) | 9 (9.4) | |
| Absence | 124 (77.5) | 37 (57.8) | 87 (90.6) | |
| Peritumoral hypo-intensity on HBP | < 0.001 | |||
| Presence | 52 (32.5) | 38 (59.4) | 14 (14.6) | |
| Absence | 108 (67.5) | 26 (40.6) | 82 (85.4) | |
| Diffusion parameters | ||||
| ADC | 1.11 ± 0.25 | 1.03 ± 0.23 | 1.16 ± 0.25 | < 0.001 |
| D | 0.89 ± 0.24 | 0.78 ± 0.22 | 0.96 ± 0.22 | < 0.001 |
| D* | 21.3 ± 11.1 | 23.4 ± 12.3 | 20.0 ± 10.1 | 0.057 |
| f | 28.1 ± 12.8 | 29.2 ± 15.8 | 27.4 ± 10.3 | 0.378 |
Unless otherwise specified, data are number of patients with percentage in parentheses.
Categorical variables were compared by using the χ2 test or Fisher exact test.
LI-RADS, Liver Imaging Reporting and Data System; AP, arterial phase; HBP, hepatobiliary phase; ADC, apparent diffusion coefficient; D, true diffusion coefficient; D* , pseudo-diffusion coefficient; f, perfusion fraction;
3.4. Univariate and multivariate analyses of the risk factors of recurrence
At univariate analysis, AFP > 400 ng/ml (HR, 1.762; 95 % CI, 1.075–2.889; p < 0.05), tumor size (HR, 2.231; 95 % CI, 1.669–2.701; p < 0.001), nodule-in-nodule architecture (HR, 2.003; 95 % CI, 1.020–3.936; p < 0.05), mosaic architecture (HR, 3.476; 95 % CI, 2.122–5.694; p < 0.001), non-smooth tumor margin (HR, 3.403; 95 % CI, 1.906–6.073; p < 0.001), intratumor necrosis (HR, 2.542; 95 % CI, 1.549–4.169; p < 0.001), satellite nodule (HR, 0.282; 95 % CI, 0.171–0.464; p < 0.001), peritumoral hypo-intensity on HBP (HR, 5.689; 95 % CI, 3.395–9.533; p < 0.001), ADC (HR, 0.629; 95 % CI, 0.479–0.827; p < 0.01) and D (HR, 0.538; 95 % CI, 0.410–0.705; p < 0.001) were found to be statistically significant with early recurrence. At multivariate analysis, tumor size (HR, 1.435; 95 % CI, 0.702–2.026; p < 0.05), mosaic architecture (HR, 0.790; 95 % CI, 0.421–1.480; p < 0.05), non-smooth tumor margin (HR, 1.775; 95 % CI, 0.941–3.273; p < 0.05), intratumor necrosis (HR, 1.414; 95 % CI, 0.807–2.476; p < 0.05), satellite nodule (HR, 0.648; 95 % CI, 0.352–1.191; p < 0.01), peritumoral hypo-intensity on HBP (HR, 2.786; 95 % CI, 1.141–6.802; p < 0.05) and D (HR, 0.658; 95 % CI,0.487–0.889; p < 0.01) were the independent risk factors for recurrence. The detailed results of the univariate and multivariate analyses are presented in Table 3.
Table 3.
Univariate and multivariate analyses of independent risk factors associated with early HCC recurrence in the training cohort.
| Variables | Univariate analysis | Multivariate analysis | ||
|---|---|---|---|---|
| HR (95 % CI) | p value | HR (95 % CI) | p value | |
| Age | 1.026 (0.994–1.059) | 0.115 | ||
| Gender (male vs female) | 0.857 (0.466–1.576) | 0.620 | ||
| Etiology (HBV vs others) | 1.173 (0.368–3.738) | 0.788 | ||
| Cirrhosis (presence vs absence) | 1.287 (0.516–3.206) | 0.588 | ||
| AFP > 400 ng/ml | 1.762 (1.075–2.889) | 0.027 | 1.193 (0.702–2.026) | 0.514 |
| Child-Pugh score | 1.027 (0.142–7.410) | 0.979 | ||
| Tumor size (cm) | 2.213 (1.669–2.701) | < 0.001 | 1.435 (1.041–1.977) | 0.027 |
| Non-rim hyperenhancement on AP | 1.141 (0.358–3.635) | 0.824 | ||
| Non-peripheral washout | 0.914 (0.367–2.278) | 0.848 | ||
| Intratumor hemorrhage | 1.151 (0.601–2.204) | 0.671 | ||
| Nodule-in-nodule architecture | 2.003 (1.020–3.936) | 0.063 | 1.554 (0.721–3.347) | 0.260 |
| Corona enhancement | 1.154 (0.655–2.032) | 0.620 | ||
| Mosaic architecture | 3.476 (2.122–5.694) | < 0.001 | 0.790 (0.421–1.480) | 0.046 |
| Non-smooth tumor margin | 3.403 (1.906–6.073) | < 0.001 | 1.755 (0.941–3.273) | 0.007 |
| Intratumor necrosis | 2.542 (1.549–4.169) | < 0.001 | 1.414 (0.807–2.476) | 0.022 |
| Satellite nodule | 0.282 (0.171–0.464) | < 0.001 | 0.648 (0.352–1.191) | 0.026 |
| Peritumoral hypo-intensity on HBP | 5.689 (3.395–9.533) | < 0.001 | 2.786 (1.141–6.802) | < 0.001 |
| ADCa | 0.629 (0.479–0.827) | 0.001 | ||
| Da | 0.538 (0.410–0.705) | < 0.001 | 0.658 (0.487–0.889) | 0.006 |
| D*a | 1.195 (0.952–1.502) | 0.125 | ||
| fa | 1.067 (0.831–1.370) | 0.609 |
Variables with p < 0.05 at univariate analysis were applied to multivariate analysis using a cox regression model.
HBV, Hepatitis B virus; AFP, alpha-fetoprotein; HR, hazard ratio; CI, confidence intervals; AP, arterial phase; HBP, hepatobiliary phase; ADC, apparent diffusion coefficient; D, true diffusion coefficient; D* , pseudo-diffusion coefficient; f, perfusion fraction;
aData at univariate and multivariate logistic analyses were transformed into a z normalization
3.5. Development and validation of the nomogram for prediction occurrence
Using the results from multivariate logistic regression, a nomogram for predicting early recurrence was developed. The model incorporated seven variables: six MRI features (tumor size, mosaic architecture, non-smooth tumor margin, intratumor necrosis, satellite nodule, and peritumoral hypo-intensity on HBP), and one diffusion parameter (D value). The nomogram exhibited excellent predictive performance, with a concordance index (C-index) of 0.913 in the training cohort and 0.875 in the validation cohort (Fig. 3).
Fig. 3.
Nomogram(a) integrated radiological features and D value for predicting recurrence probabilities in training cohort. Calibration curves for the nomogram in training (b) and validation cohort (c). Calibration curves depict the agreement of the model between the predicted risks of early recurrence and the actual observed recurrence. The apparent line and bias-corrected line are close to ideal dashed line in both training (C index, 0.913) and validation cohort (C index, 0.875), represent a high prediction accuracy of the model.
3.6. Imaging risk factors and nomogram-predicted risk stratification for RFS
All the identified factors were significantly correlated with early RFS (all p < 0.001), as illustrated in Fig. 4a–g. In addition, risk scores were calculated for each patient based on nomogram with the training cohort ranging from 10 to 308 and the validation cohort ranging from 12 to 303. Patients were stratified into low- and high-risk groups based on whether their scores fell below or above the median score, respectively. Nomogram-predicted high-risk patients exhibited significantly higher recurrence rates compared to low-risk patients in both the training cohort and the validation cohort (all p < 0.001, Fig. 4h, i, respectively)
Fig. 4.
Kaplan-Meier curves of imaging features, including tumor diameter (a), mosaic architecture (b), tumor margin (c), intratumor necrosis (d), satellite nodule (e), peritumoral hypo-intensity on HBP (f) and D value (g) for predicting early recurrence free survival. After the nomogram was developed, each patient was calculated a risk score based on the nomogram and categorized into high and low risk groups. The prognostic value in the training cohort (h) and the validation cohort (i) were also calculated. The P values were calculated using the log-rank test.
4. Discussion
In this study, we identified six clinical indicators, eleven imaging features, and four diffusion parameters as potential biomarkers for predicting early recurrence of HCC after resection. Among these, six imaging features on gadoxetic acid-enhanced MRI and D value were determined to be independent risk factors for early recurrence. By combining these factors, we developed a predictive model that demonstrated strong efficacy in both the training and validation cohorts.
Previous studies have reported that early recurrence is closely associated with tumor differentiation and may stem from microscopic metastases in the liver, such as MVI and VETC [20]. These pathological features, correspond to specific biological behaviors of the tumor that influence its microstructural and hemodynamic characteristics that can be indirectly assessed through imaging. In this study, we found that larger tumor size, mosaic structure, intra-tumor necrosis, non-smooth tumor margin, peritumoral hypo-intensity on HBP and D-value were all independent risk factors for predicting early recurrence of HCC. These imaging features are reflective of tumor biological behavior. Specifically, a larger tumor diameter indicates more aggressive tumor biology, while mosaic structure and intra-tumor necrosis are commonly observed in larger tumors, often associated with tumor hypoxia and ischemia, which also reflect tumor heterogeneity [21], [22]. Non-smooth tumor margin often results from extra-nodular growth or the fusion of multiple nodules, reflecting irregular growth patterns [23]. The presence of peritumoral hypo-intensity on HBP correlated significantly with MVI, it could be attributed to the dysfunction of organic anion-transporting polypeptide transporters of hepatocytes around the tumor, which are a consequence of hemodynamic changes associated with the obstruction of micro portal veins by MVI [24], [25], [26]. Moreover, Additionally, the D value, which has been shown to outperform the apparent diffusion coefficient (ADC) in predicting tumor differentiation and MVI status, demonstrated a stronger correlation with early recurrence in our study [27].
Although numerous studies have established a relationship between certain imaging features and early recurrence, these results remain somewhat controversial and insufficient for guiding individualized treatment [28]. This is partly due to the fact that feature determination often relies on subjective interpretation, which limits the specificity and sensitivity of individual features. In our study, while multiple risk factors were identified, the prevalence of these factors within the overall cohort was relatively low; for example, mosaic architecture and satellite nodules were present in only 36 % (58/160) and 23 % (36/160), respectively. To address these limitations, we integrated all identified risk factors into a comprehensive prediction model, which facilitates more accurate individual risk stratification. Moreover, compared to previous studies, our model incorporates quantitative diffusion parameters, thus integrating both subjective and objective factors and reducing potential biases introduced by subjective analysis [29], [30]. Although previous studies have suggested that diffusion kurtosis imaging (DKI) parameters can also predict HCC recurrence, DKI requires higher b-values, specific field strength, and longer scan times, which make it less suitable for routine clinical practice [31]. In contrast, IVIM imaging, as used in our study, is more broadly applicable across scanner types and offers a practical alternative for clinical implementation. In summary, the prediction model developed in this study demonstrated excellent performance in both the training and validation cohorts, outperforming previous models [32], [33]. This model can be utilized to identify patients at high risk of recurrence, allowing for personalized treatment strategies such as expanded resection, neoadjuvant therapy, or immunotherapy to reduce the risk of postoperative recurrence. Additionally, it support the development of more intensive follow-up protocols, or the use of more sensitive follow-up tools, to improve early detection of recurrence.
Our study has several limitations. First, as a retrospective study, selection bias may be present in both the training and validation cohorts. Second, while external validation was conducted, both cohorts were scanned using 3.0 T scanners, which may limit the generalizability of the findings. Third, the majority of patients in our cohort were HBV-infected, which may restrict the applicability of our findings to patients with non-HBV-related HCC. Fourth, we focused on liver-specific contrast agents, so the conclusions may not apply to studies using extracellular contrast agents. Fifth, postoperative monitoring intervals ranged from 3 to 6 months, and while this variability is based on clinical recommendations, it may influence the accuracy of recurrence detection. Finally, the assessment of imaging features relied on subjective analysis, and while reproducibility could be improved with the aid of artificial intelligence, this remains an area for future enhancement [34].
In conclusion, our findings suggest that radiologic features, including tumor size, non-smooth tumor margin, mosaic architecture, intra-tumor necrosis, peritumoral hypo-intensity on HBP, and the D value, can be used to predict early recurrence of HCC. The integration of these prognostic features into a nomogram provides robust performance in both internal and external cohorts, offering potential for personalized treatment strategies and improving outcomes for patients with HCC.
Funding statement
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Informed consent
Informed consent was obtained from all individual participants included in the study.
Ethical statement
All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
CRediT authorship contribution statement
Guo Da: Writing – original draft, Software. Jin Yu: Validation, Supervision, Project administration, Conceptualization. Liu Liping: Writing – review & editing, Methodology.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Footnotes
Supplementary data associated with this article can be found in the online version at doi:10.1016/j.ejro.2025.100643.
Appendix A. Supplementary material
Supplementary material
Supplementary material
References
- 1.Toh M.R., Wong E.Y.T., Wong S.H., et al. Global epidemiology and genetics of hepatocellular carcinoma. Gastroenterology. 2023;164(5):766–782. doi: 10.1053/j.gastro.2023.01.033. [DOI] [PubMed] [Google Scholar]
- 2.Singal A.G., Kanwal F., Llovet J.M. Global trends in hepatocellular carcinoma epidemiology: implications for screening, prevention and therapy. Nat. Rev. Clin. Oncol. 2023;20(12):864–884. doi: 10.1038/s41571-023-00825-3. [DOI] [PubMed] [Google Scholar]
- 3.Ganesan P., Kulik L.M. Hepatocellular carcinoma: new developments. Clin. Liver Dis. 2023;27(1):85–102. doi: 10.1016/j.cld.2022.08.004. [DOI] [PubMed] [Google Scholar]
- 4.Brown Z.J., Tsilimigras D.I., Ruff S.M., et al. Management of hepatocellular carcinoma: a review. JAMA Surg. 2023;158(4):410–420. doi: 10.1001/jamasurg.2022.7989. [DOI] [PubMed] [Google Scholar]
- 5.Nevola R., Ruocco R., Criscuolo L., et al. Predictors of early and late hepatocellular carcinoma recurrence. World J. Gastroenterol. 2023;29(8):1243–1260. doi: 10.3748/wjg.v29.i8.1243. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Isik B., Gonultas F., Sahin T., et al. Microvascular venous invasion in hepatocellular carcinoma: why do recurrences occur? J. Gastrointest. Cancer. 2020;51(4):1133–1136. doi: 10.1007/s12029-020-00487-9. [DOI] [PubMed] [Google Scholar]
- 7.Wang M., Cao L., Wang Y., et al. The prognostic value of vessels encapsulating tumor clusters (VETC) in patients with hepatocellular carcinoma: a systematic review and meta-analysis. Clin. Transl. Oncol. 2024;26(8):2037–2046. doi: 10.1007/s12094-024-03427-2. [DOI] [PubMed] [Google Scholar]
- 8.Hwang S.H., Rhee H. Radiologic features of hepatocellular carcinoma related to prognosis. J. Liver Cancer. 2023;23(1):143–156. doi: 10.17998/jlc.2023.02.16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Jiang H., Qin Y., Wei H., et al. Prognostic MRI features to predict postresection survivals for very early to intermediate stage hepatocellular carcinoma. Eur. Radio.. 2024;34(5):3163–3182. doi: 10.1007/s00330-023-10279-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Ichikawa S., Goshima S. Gadoxetic acid-enhanced liver MRI: everything you need to know. Invest Radio.. 2024;59(1):53–68. doi: 10.1097/RLI.0000000000000990. [DOI] [PubMed] [Google Scholar]
- 11.Jiang H., Wei J., Fu F., et al. Predicting microvascular invasion in hepatocellular carcinoma: a dual-institution study on gadoxetate disodium-enhanced MRI. Liver Int. 2022;42(5):1158–1172. doi: 10.1111/liv.15231. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Hong S.B., Choi S.H., Kim S.Y., et al. MRI features for predicting microvascular invasion of hepatocellular carcinoma: a systematic review and meta-analysis. Liver Cancer. 2021;10(1):10–24. doi: 10.1159/000513704. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Chen F.M., Du M., Qi X., et al. Nomogram estimating vessels encapsulating tumor clusters in hepatocellular carcinoma from preoperative gadoxetate disodium-enhanced MRI. J. Magn. Reson. Imaging. 2023;57(6):1893–1905. doi: 10.1002/jmri.28488. [DOI] [PubMed] [Google Scholar]
- 14.Le Bihan D. What can we see with IVIM MRI? Neuroimage. 2019;187(2):56–67. doi: 10.1016/j.neuroimage.2017.12.062. [DOI] [PubMed] [Google Scholar]
- 15.Zhou Y., Zheng J., Yang C., et al. Application of intravoxel incoherent motion diffusion-weighted imaging in hepatocellular carcinoma. World J. Gastroenterol. 2022;28(27):3334–3345. doi: 10.3748/wjg.v28.i27.3334. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Min J.H., Lee M.W., Park H.S., et al. Interobserver variability and diagnostic performance of gadoxetic acid-enhanced MRI for predicting microvascular invasion in hepatocellular carcinoma. Radiology. 2020;297(3):573–581. doi: 10.1148/radiol.2020201940. [DOI] [PubMed] [Google Scholar]
- 17.Wei H., Jiang H., Zheng T., et al. LI-RADS category 5 hepatocellular carcinoma: preoperative gadoxetic acid-enhanced MRI for early recurrence risk stratification after curative resection. Eur. Radiol. 2021;31(4):2289–2302. doi: 10.1007/s00330-020-07303-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Sheng Y., Wang Q., Liu H.F., et al. Preoperative nomogram incorporating clinical factors, serological markers and LI-RADS MRI features to predict early recurrence of hepatocellular carcinoma treated with transarterial chemoembolization. Acad. Radiol. 2023;30(7):1288–1297. doi: 10.1016/j.acra.2022.10.020. [DOI] [PubMed] [Google Scholar]
- 19.Wang Q., Yu G., Qiu J., et al. Application of intravoxel incoherent motion in clinical liver imaging: a literature review. J. Magn. Reson. Imaging. 2024;60(2):417–440. doi: 10.1002/jmri.29086. [DOI] [PubMed] [Google Scholar]
- 20.Fuster-Anglada C., Mauro E., Ferrer-Fàbrega J., et al. Histological predictors of aggressive recurrence of hepatocellular carcinoma after liver resection. J. Hepatol. 2024;81(6):995–1004. doi: 10.1016/j.jhep.2024.06.018. [DOI] [PubMed] [Google Scholar]
- 21.Usta S., Kayaalp C. Tumor diameter for hepatocellular carcinoma: why should size matter? J. Gastrointest. Cancer. 2020;51(4):1114–1117. doi: 10.1007/s12029-020-00483-z. [DOI] [PubMed] [Google Scholar]
- 22.Öcal O., Ingrisch M., Ümütlü M.R., et al. Prognostic value of baseline imaging and clinical features in patients with advanced hepatocellular carcinoma. Br. J. Cancer. 2022;126(2):211–218. doi: 10.1038/s41416-021-01577-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Song L., Li J., Luo Y. The importance of a nonsmooth tumor margin and incomplete tumor capsule in predicting HCC microvascular invasion on preoperative imaging examination: a systematic review and meta-analysis. Clin. Imaging. 2021;76(8):77–82. doi: 10.1016/j.clinimag.2020.11.057. [DOI] [PubMed] [Google Scholar]
- 24.Lee S., Kim K.W., Jeong W.K., et al. Gadoxetic acid-enhanced MRI as a predictor of recurrence of HCC after liver transplantation. Eur. Radiol. 2020;30(2):987–995. doi: 10.1007/s00330-019-06424-0. [DOI] [PubMed] [Google Scholar]
- 25.Bae J.S., Kim J.H., Lee D.H., et al. Hepatobiliary phase of gadoxetic acid-enhanced MRI in patients with HCC: prognostic features before resection, ablation, or TACE. Eur. Radiol. 2021;31(6):3627–3637. doi: 10.1007/s00330-020-07499-w. [DOI] [PubMed] [Google Scholar]
- 26.Chen H., Dong H., He R., et al. Optimizing predictions: improved performance of preoperative gadobenate-enhanced MRI hepatobiliary phase features in predicting vessels encapsulating tumor clusters in hepatocellular carcinoma-a multicenter study. Abdom. Radiol. 2024;49(10):3412–3426. doi: 10.1007/s00261-024-04283-y. [DOI] [PubMed] [Google Scholar]
- 27.Zhou L., Qu Y., Quan G., et al. Nomogram for predicting microvascular invasion in hepatocellular carcinoma using gadoxetic acid-enhanced MRI and intravoxel incoherent motion imaging. Acad. Radiol. 2024;31(2):457–466. doi: 10.1016/j.acra.2023.06.028. [DOI] [PubMed] [Google Scholar]
- 28.Deng Y., Yang D., Xu H., et al. Diagnostic performance of imaging features in the HBP of gadoxetate disodium-enhanced MRI for microvascular invasion in hepatocellular carcinoma: a meta-analysis. Acta Radiol. 2022;63(10):1303–1314. doi: 10.1177/02841851211038806. [DOI] [PubMed] [Google Scholar]
- 29.Lu M., Qu Q., Xu L., et al. Prediction for aggressiveness and postoperative recurrence of hepatocellular carcinoma using gadoxetic acid-enhanced magnetic resonance imaging. Acad. Radiol. 2023;30(5):841–852. doi: 10.1016/j.acra.2022. [DOI] [PubMed] [Google Scholar]
- 30.Wang L., Feng B., Li D., et al. Risk stratification of solitary hepatocellular carcinoma ≤ 5 cm without microvascular invasion: prognostic values of MR imaging features based on LI-RADS and clinical parameters. Eur. Radiol. 2023;33(5):3592–3603. doi: 10.1007/s00330-023-09484-5. [DOI] [PubMed] [Google Scholar]
- 31.Cao X., Shi H., Dou W.Q., et al. Can DKI-MRI predict recurrence and invasion of peritumoral zone of hepatocellular carcinoma after transcatheter arterial chemoembolization? World J. Gastrointest. Surg. 2022;14(10):1150–1160. doi: 10.4240/wjgs.v14.i10.1150. 27. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Chan A.W.H., Zhong J., Berhane S., et al. Development of pre and post-operative models to predict early recurrence of hepatocellular carcinoma after surgical resection. J. Hepatol. 2018;69(6):1284–1293. doi: 10.1016/j.jhep.2018.08.027. [DOI] [PubMed] [Google Scholar]
- 33.Wei H., Jiang H., Qin Y., et al. Comparison of a preoperative MR-based recurrence risk score versus the postoperative score and four clinical staging systems in hepatocellular carcinoma: a retrospective cohort study. Eur. Radiol. 2022;32(11):7578–7589. doi: 10.1007/s00330-022-08811-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Xia T., Zhao B., Li B., et al. MRI-based radiomics and deep learning in biological characteristics and prognosis of hepatocellular carcinoma: opportunities and challenges. J. Magn. Reson Imaging. 2024;59(3):767–783. doi: 10.1002/jmri.28982. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary material
Supplementary material