Evaluating the severity of microvascular invasion in hepatocellular carcinoma, by probing the combination of enhancement modes and growth patterns through magnetic resonance imaging

Hepatocellular carcinoma (HCC) is one of the top lethal malignancies worldwide with the largest HCC burden in our country, even if new cases are diagnosed and cancer-related deaths are yearly reduced.¹ Though steady advances in screening and clinical strategies improve early detection and systemic therapies, the survivals remain undesirable due to the high incidence of recurrence and metastasis.^1,2 Microvascular invasion (MVI) should be one of the primary responsibilities for long-term survival and an explicit risk factor for relapse.³ According to the findings^4,5, not only the presence of MVI, in particular, severity grades closely correlate with long-term recurrence and cause varying prognoses. In general, frequent tests and intense interventions are necessary for patients’ duration, however, they are not indicated for all grades of MVI. Research showed that compared with low-risk MVI, high-risk patients sustain a higher rate of recurrence of 86.1% and a lower rate of overall survival of 67% in 3-year postoperative.^4,6 Therefore, patients who have severe MVI stage require more aggressive treatment procedures and sophisticated follow-up protocols, that theoretically and effectively reduce the rate of recurrence after treatment and favor prognosis.^7,8 To assess the severity of MVI, accurate detection at several points is needed in postoperative pathological specimens and is hardly applied in clinical practice before surgery, due to unreliably diagnosed with limited details and some repulsive drawbacks.^8,9 Besides, no non-invasive option exists currently that can surpass the “gold standard” of histological specimens in MVI detection or categorization.¹⁰

Thanks to the inclusion of sequences and parameters, magnetic resonance imaging (MRI) has superiorities in detecting and evaluating the status of MVI, which compensates for the absence of preoperative assessment by biopsy, in which hepatobiliary contrast agents play an essential role as well^11-13, it increased the sensitivity of diagnosing small HCCs that depicted non-arterial enhancing mode.¹⁴ Gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (GD-EOB-DTPA) enhanced MRI has been reported as useful for assessing the microstructure and particular signs of the lesion.¹⁵ Hitherto, certain imaging features have been defined as characteristic of the MVI-positive, and a portion of the literature has described and explored MVI stratification, however, no consensus has been achieved, and no distinctive imaging findings can be graded to determine the severity of MVI.^12,16 Thus preoperative assessment of MVI status is required in selecting the appropriate surgical method and establishing suitable therapy in different durations.¹⁷ Although radiomics offers several benefits, the operations are relatively complex, and the predictive results are indirect, lacking validation, and unsuitable for routine clinics which unequipped with specialized machines and software.

That is to say, a reusable, application-friendly noninvasive test and predictive model need to be developed to identify the severity of MVI and substantial advantages for patients in the selection of preoperative surgical procedures.¹⁸ Nevertheless, no research has considered a combination of tumor growth patterns and enhancement modes to develop a nomogram model that can predict MVI grade. Hence, we extract clinicoradiological factors and imaging descriptions that can be clinically valuable for detecting MVI existence and its grade and establish a fusion model for preoperative evaluation.

MVI occurrence indicates aggressive tumor biology and is attributed to an unfavorable prognosis. Even with a methodical therapy, evaluation of MVI grading continues to be a challenge and affects personalized care.²³ However, most researchers simply utilized a binary classification method to predict the presence of MVI, without considering MVI grading, particularly M2 grade. In terms of the prognosis for HCC, M2 status is more likely to experience recurrence, intrahepatic metastases, and residual tumor than M1 status.²⁴ Therefore, we extract enhanced modes and growth patterns of tumors from MRI to provide a universal, noninvasive, and robust method for preoperative patient prediction of MVI status. In our study, accuracy for M2 grade diagnosis is up to 79%, with corresponding sensitivity and specificity of 74% and 82%, respectively.

Considering earlier publications²⁵, serum AFP level was significant risk factor for MVI existence and are associated with factors of tumor aggressiveness, including early recurrence.²⁶ In this study, the serum AFP level was significantly related to MVI occurrence, even when the AFP value was slightly above the limit, but without significant statistics in predicting M2 grade. The fact that AFP is not a particular marker for HCC and can remain unstable under a range of circumstances could be the cause. For Chen’s results, HCCs with M1 or M2 grade may have different high AFP level and lack significance⁶, which matches our findings. Based on the Milan criteria, a tumor greater than 5 cm is ineligible for liver transplantation, however, MVI-positive HCCs less than 5 cm may lose the second chance in transplant and get worse outcomes. Nonetheless, varying research produced tumor size in diverse critical values when assessing the prediction of MVI. Most researchers proposed a tumor size greater than 5.0 cm as a general visual predictor and MVI status positively correlates with tumor size.⁸ According to Shirabe et al.²⁷ a tumor size exceeding 3.6 cm was considered to be a reliable indicator of MVI. In our search, a tumor size larger than 3.9 cm was regarded to be a sign of MVI. In this study, it is worth considering that the P value of HBeAg was close to 0.05 in univariate analysis, and in Jian Liu’s study²⁸, patients with HBeAg positive were an independent factor of MVI-positive and associated with a poorer prognosis. Given the similarity of our results, we suppose that positive HBeAg will probably be considered as a potential factor for assessing the M2 grade with further sample enlargement.

The previous study indicates lesions expressed in morphological changes are more easily discovered and acceptable than other complex enhancement and signal features in clinical diagnosis.²⁶ Adjacent non-tumor tissue was thought to be the region where HCC cells frequently spread into the surrounding liver tissues, causing uneven tumor surface and irregular growth⁸, implying the occurrence of MVI, which was consistent with our findings. Li Y et al.²⁹ applied the tumor surface inclination angles to assess HCCs with MVI presence, which had a higher surface irregularity rate and worse sphericity. In this study, we utilized the Max/Min-R as a substitute for flatness, as done in Zheng’s⁸ study, to serve as an easily implemented assessment method for tumor irregularity instead of preoperative 3D morphological evaluation. The latter cannot be routinely performed in the clinic. Our research further refined and categorized the marginal morphological features to evaluate the distinctive growth pattern of M2.^20,29 Our findings indicate that a Max/Min-R exceeding 1.22 suggests a more severe irregularity and is substantially correlated with a high-risk factor for M2, which aligns with the results reported in Zheng’s study. Among the MVI-positive nomogram predictors, the immersed growth had the highest score (100 points on the scale axis). For the M2 status nomogram, confluent multinodule and immersed growth had a higher score for M2 status. Wang X et al.³⁰ illustrate that HCC with irregular shape or rough edge indicating multiple microscopic layers budding at the periphery of the lesion is more prone to induce tumors to grow in an immersed or multinodular manner.

Among the image features that were repeatedly verified and recognized, some specific features of the peritumoral region corroborate the MVI presence, including arterial rim-enhancement, arterial peritumoral enhancement on AP, and peritumoral hypointensity on HBP. The aforementioned features of the tumor-liver region may suggest abnormal hemodynamics in adjacent tissues caused by extracellular matrix disarray, tissue distortion, tumor cell blockage, cell proliferation, and inflammation.^8,31 These criteria are significant in differentiating MVI-positive in clinical importance and align with our findings. Previous research has categorized the enhancement modalities for HCC into typical and atypical patterns in assessing MVI occurrence.^32–35 However, they may not fully distinguish between M1 and M2 as identified in the research.

Intratumor vascularity is frequently utilized as an imaging marker for aggressive HCC and displays various enhanced patterns, and some studies have been reported to be associated with MVI occurrence, early relapse, and recurrencefree survival.^33,36–38 Hence, we can assess indirectly inner-tumor vessel clusters by observing the enhancement pattern. In this study, we divided the enhancement patterns into four categories, and MVI-positive HCCs are more likely to exhibit the no/minimal enhancement pattern, which could indicate hypoperfusion as a result of tumor cell blockage in the peritumoral area of the microvas-culature.³⁹ Likewise, the promising results from Li Zhang et al.³⁶ analyzed the triphasic CT enhancement images by hemodynamics software and showed that the presence of MVI had lower hemodynamic perfusion parameters, which parallels what we observed. While HCCs with M2 grade exhibit a higher propensity for developing persistent enhanced mode and are unlikely to have washin/washout enhanced mode. This may be attributed to the existence of vessels encapsulating tumor clusters, or indirectly resulting in persistent enhancement.³³ Or maybe peri-tumoral microvascular branches were disrupted or obstructed by the tumor cells and caused compensatory hyperperfusion in subsequent phases.⁸

By adding tumor morphology and enhancement modes to the nomogram, we improved our study’s clinical applicability while raising the MVI grading accuracy. Yumeng Li et al developed an MVI prediction model based on integrating laboratory indicators and 3D morphological features, yielding an AUC of 0.831²⁹, which could potentially improve the predictive efficiency. Although our study used two-dimensional morphological features to assess tumor growth patterns, the nomogram showed satisfactory performance in identifying the presence of MVI and its M2 status. Based on the calibration curves, both nomograms showed good consistency between the observed probability and the predicted probability. Further evidence of the two nomograms with great clinical value came from DCA, which demonstrated a net benefit of the nomograms with a risk threshold greater than 0.15. In addition, radiologists or surgeons can quickly sense the features and scores from these in clinical practice, even if the variations in scanning parameters and apparatus.

Our study skipped over the clinical imaging features used for M1 status or developing a model to differentiate it. Due to the existence of similar issues in other models^8,40, standardizing the clinical model of M1 was not so functional. We also omitted pathology features from the multivariate analysis since they depend on an invasive exam. Adding histopathologic features would have compromised the model’s validity and financial viability. Despite their potential, radiomics and deep learning both have high labor requirements and poor interpretability; they also require numerous samples, higher-quality imaging, and scanning conditions.⁴¹ Most institutes lack data processors and technicians to process. Compared to our nomogram model, it is not as practical and convenient for clinical use.

Several limitations of our study should be noted. Initially, there can be selection bias due to the retrospective design of this research. Additionally, the patient sample size was sufficient, but somewhat in certain groups, the amount was limited due to the multi-characteristics and the dispersed distribution of the cases, although we emerged the patients from two centers and discarded the group validation. Furthermore, our study solely focused on analyzing the appearance of tumor morphology and enhancement; in contrast, the utilization and analysis merger of laboratory tests (AFP, PIVKA-II) in predicting MVI grades were weakly relevant. Lastly, there is no concrete evidence establishing a direct correlation between radiological features and MVI grades. Future large dataset integration, multi-parameter combining, prospective multicentre trials, and independent validation would be required to investigate the relationship between radiological features and MVI grade.

In conclusion, we incorporated easy-to-recognize features and established nomograms to enable satisfying MVI status prediction. When clinicians consider factors and score the nomogram systems to deem the patient at high risk of MVI or maybe categorized as M2 grade, monitoring plan selection and personalized therapy design are helpful for HCC patients to start treatment in an efficient and curative way.