New frontiers in hepatocellular carcinoma: Precision imaging for microvascular invasion prediction

Abstract

This paper highlights the innovative approach and findings of the recently published study by Xu et al, which underscores the integration of radiomics and clinicoradiological factors to enhance the preoperative prediction of microvascular invasion in patients with hepatitis B virus-related hepatocellular carcinoma (HBV-HCC). The study’s use of contrast-enhanced computed tomography radiomics to construct predictive models offers a significant advancement in the surgical planning and management of HBV-HCC, potentially transforming patient outcomes through more personalized treatment strategies. This editorial commends the study's contribution to precision medicine and discusses its implications for future research and clinical practice.

Key Words: Microvascular invasion prediction; Hepatocellular carcinoma; Radiomics; Precision imaging; Hepatitis B virus

Core Tip: The integration of precision medicine in hepatocellular carcinoma (HCC) management is crucial for tailoring treatment to individual patient characteristics, and leveraging radiomics serves as a powerful non-invasive tool for predicting microvascular invasion preoperatively, thereby guiding more informed surgical decisions. To enhance the robustness and applicability of predictive models, multicentric studies involving diverse populations should be promoted, alongside the integration of radiomics with genetic and molecular markers for a more comprehensive understanding of the tumor microenvironment. Embracing advancements in imaging technologies and conducting cost-effectiveness analyses are essential for justifying the adoption of radiomics in clinical practice. Additionally, addressing ethical considerations regarding patient data privacy and promoting the use of radiomics in prospective clinical trials can help validate their effectiveness in real-world settings. Investing in training for healthcare professionals will improve their interpretation of radiomics data, facilitating its routine use, while fostering collaboration among oncologists, radiologists, data scientists, and researchers will continually refine predictive models and enhance their utility in managing HCC.

TO THE EDITOR

We read with great interest the article by Xu et al[1], titled "Evaluating microvascular invasion in hepatitis B virus-related hepatocellular carcinoma based on contrast-enhanced computed tomography radiomics and clinicoradiological factors" published recently in World Journal of Gastroenterology. This study evaluated microvascular invasion (MVI) in hepatocellular carcinoma (HCC) by developing a predictive model that integrated three key independent factors: Non-smooth tumor margin [odds ratio (OR) = 2.655], absence of peritumoral hypointensity ring (OR = 5.360), and a radiomics score based on delayed-phase computed tomography (CT) images (OR = 6.552). The model achieved promising results with an area under the curve (AUC) of 0.841 and 0.768 in training and validation datasets, respectively. When contextualizing within existing research, previous approaches have ranged from Beaufrère et al's genetic 6-gene signature to various imaging-based studies like Lei et al's clinicoradiological nomogram and CT-based radiomics model[2,3]. Xu et al's work stands out through several innovations: It specifically targeted hepatitis B virus-related HCC (HBV-related HCC), utilized more accessible CT imaging rather than magnetic resonance imaging (MRI), incorporated both tumor and peritumoral features, and uniquely addressed both MVI prediction and M2 (high-risk) identification[1]. The study's methodological strengths included rigorous validation approaches, sensitivity analyses, and practical clinical applicability demonstrated through decision curve analysis. The key performance metrics from Xu et al’s study are indicated in Tables 1, 2, and 3[1]. The significance of their work cannot be overstated, as it marks a pivotal advancement in the non-invasive prediction of MVI, which is crucial for surgical planning and prognosis in HCC patients.

Table 1 Microvascular invasion status prediction model performance[1].

Dataset	AUC (95%CI)	Sensitivity (%)	Specificity (%)	Accuracy (%)	Calibration (χ²)	P value
Training (n = 189)	0.841 (0.783-0.898)	88.8	64.4	79.4	8.327	0.402
Validation (n = 81)	0.768 (0.664-0.872)	80.9	61.8	72.8	12.804	0.119

AUC: Area under the curve. Citation: Xu ZL, Qian GX, Li YH, Lu JL, Wei MT, Bu XY, Ge YS, Cheng Y, Jia WD. Evaluating microvascular invasion in hepatitis B virus-related hepatocellular carcinoma based on contrast-enhanced computed tomography radiomics and clinicoradiological factors. World J Gastroenterol 2024; 30: 4801-4816.

Table 2 High-risk group status prediction model performance[1].

Dataset	AUC (95%CI)	Sensitivity (%)	Specificity (%)	Accuracy (%)	Calibration (χ²)	P value
Training (n = 189)	0.865 (0.797-0.934)	78.6	86.4	84.7	9.311	0.317
Validation (n = 81)	0.798 (0.676-0.919)	57.1	93.3	84.0	2.280	0.131

AUC: Area under the curve. Citation: Xu ZL, Qian GX, Li YH, Lu JL, Wei MT, Bu XY, Ge YS, Cheng Y, Jia WD. Evaluating microvascular invasion in hepatitis B virus-related hepatocellular carcinoma based on contrast-enhanced computed tomography radiomics and clinicoradiological factors. World J Gastroenterol 2024; 30: 4801-4816.

Table 3 Independent risk factors for microvascular invasion and high-risk group[1].

Prediction target	Risk factor	Odds ratio (95%CI)	P value
MVI status	Non-smooth tumor margin	2.655 (1.241-5.679)	0.012
	Absence of peritumoral hypointensity ring	5.360 (1.857-15.477)	0.002
	Radscore (ROI tumor-DP)	6.552 (3.364-12.761)	< 0.001
M2 status	AFP level	3.818 (1.380-10.563)	0.01
	Enhancing capsule	3.962 (1.628-9.643)	0.002
	AST level	3.760 (1.485-9.520)	0.005
	Radscore (ROI peri-AP)	5.967 (2.609-13.649)	< 0.001

MVI: Microvascular invasion; AUC: Area under the curve; AFP: Alpha-fetoprotein; AST: Aspartate aminotransferase; ROI ^tumor-DP: Region of interest of tumor from delayed phase; ROI ^peri-AP: Region of interest of peritumoral area from arterial phase. Citation: Xu ZL, Qian GX, Li YH, Lu JL, Wei MT, Bu XY, Ge YS, Cheng Y, Jia WD. Evaluating microvascular invasion in hepatitis B virus-related hepatocellular carcinoma based on contrast-enhanced computed tomography radiomics and clinicoradiological factors. World J Gastroenterol 2024; 30: 4801-4816.

Current state of MVI prediction in HCC

The integration of precision medicine in HCC management represents a significant advancement in personalizing treatment approaches. Current research encompasses various methodologies, from genetic signatures to imaging-based studies, with each contributing unique insights to the field. Notable examples include a study by Wang et al[4] that developed and validated a prediction model for MVI based on preoperative factors, achieving an AUC of 0.812 in their validation cohort. Their model identified key predictors such as tumor size and serum α-fetoprotein levels, demonstrating the potential of combining clinical and laboratory parameters for MVI prediction. Another significant contribution comes from Li et al[5], who utilized multi-phase magnetic resonance imaging for preoperative MVI prediction, incorporating both morphological features and dynamic enhancement patterns to achieve improved diagnostic accuracy. The application of machine learning approaches, as demonstrated by Ravikulan and Rostami[6], has further enhanced our ability to predict early recurrence in HCC patients by integrating multiple data sources. These advances in MVI prediction have critical implications for determining surgical approaches and patient outcomes[7]. While existing studies have demonstrated the potential of different predictive models, ranging from traditional imaging features to advanced radiomics approaches[8], challenges in standardization and validation persist across different patient populations and clinical settings[9]. The field continues to evolve with emerging technologies and methodologies, though the need for robust validation and standardization remains paramount.

Precision medicine applications in HCC management

The concept of precision medicine in oncology has revolutionized the approach to cancer treatment by tailoring interventions based on individual patient characteristics, including genetic profiles, lifestyle factors, and environmental influences[10]. This personalized approach is particularly crucial in HCC management due to the disease's heterogeneous nature, which is influenced by various etiologies including HBV infection, dietary aflatoxin exposure, and alcohol consumption[11]. The ability to predict MVI through radiomics, as demonstrated by Xu et al[1], adds a powerful tool to the arsenal of precision diagnostics, potentially allowing clinicians to personalize surgical and therapeutic strategies effectively. Several landmark studies have demonstrated the effectiveness of this approach in HCC management. For instance, Kalinich et al[11] developed an RNA-based signature that achieved high specificity in detecting circulating tumor cells in HCC patients, enabling early detection and personalized monitoring of disease progression. In the realm of treatment selection, Park et al[12] demonstrated how radiomics and deep learning applications can guide precision therapy by predicting treatment response and identifying patients most likely to benefit from specific interventions. This personalized approach is particularly crucial in HCC management due to the disease's heterogeneous nature, which is influenced by various etiologies including HBV infection, dietary aflatoxin exposure, and alcohol consumption. The integration of radiomics into this paradigm provides a powerful tool for predicting MVI, as shown by Wakabayashi et al[13] in their comprehensive review of quantitative radiomics applications in HCC. Their analysis revealed how radiomics features can capture subtle tumor characteristics that correlate with molecular profiles and clinical outcomes. Wei et al[14] further expanded on this concept, demonstrating how radiomics can enhance current staging systems and treatment algorithms by providing more granular information about tumor biology and potential treatment response. The ability to predict MVI preoperatively through radiomics analysis represents a significant advancement in precision diagnostics, allowing for more personalized surgical and therapeutic planning that takes into account individual tumor characteristics and risk factors. Figure 1 illustrates the radiomics workflow and its application to MVI prediction.

Figure 1 Flowchart of radiomics workflow and its application to microvascular invasion prediction. CT: Computed tomography; MRI: Magnetic resonance imaging; PET: Positron emission tomography; MVI: Microvascular invasion.

Radiomics integration and clinical implementation

Radiomics technology leverages advanced computational methods to extract numerous quantitative features from medical images, identifying patterns and characteristics that may not be visible to the human eye[12,13,15]. Recent studies have demonstrated significant progress in its clinical implementation. For instance, Zhou et al[16] developed a deep learning-based radiomics model that achieved an AUC of 0.817 in predicting early recurrence of HCC after surgical resection, incorporating both imaging features and clinical parameters. This model demonstrated superior performance compared to conventional clinical predictors alone. Similarly, Kim et al[17] established a comprehensive radiomics pipeline for HCC characterization, successfully integrating data from multiple imaging modalities (CT, MRI, and positron emission tomography) to improve diagnostic accuracy and treatment planning. The workflow integration aspects were addressed by Wang et al[18], who developed a radiomics model, with a mean AUC of 0.9804 and 0.7578 in the training and validation sets, demonstrating good calibration and satisfactory discrimination to predict 5-year survival in HCC patients. These advances in imaging technology and analysis have transformed the management of HCC by providing more precise and personalized treatment approaches. The study by Xu et al[1] harnesses this technology to evaluate features from contrast-enhanced computed tomography images that correlate with the presence of MVI in patients with HBV-related HCC[14]. By integrating these radiomics features with clinicoradiological factors into predictive models, the authors achieved impressive validation results, indicative of the high potential of this approach in clinical settings. The predictive models developed by Xu et al[1] could potentially allow surgeons to modify surgical plans, opt for wider resection margins, or even alter the approach from partial hepatectomy to transplantation in high-risk cases, thereby improving outcomes.

Study limitations and challenges

While Xu et al's study represents a significant advance in MVI prediction, several important limitations must be acknowledged[1]. The single-center retrospective design with a modest sample size of 270 patients raises concerns about selection bias and generalizability. Furthermore, the model's performance drop between training (AUC = 0.841, sensitivity 88.8%, specificity 64.4%) and validation datasets (AUC = 0.768, sensitivity 80.9%, specificity 61.8%) suggests potential overfitting issues. The complex radiomics feature extraction process and manual tumor segmentation introduce operator-dependent variability, and the moderate specificity could lead to unnecessary clinical interventions. Implementation barriers include the need for specialized radiomics expertise and time-consuming manual segmentation, which may hinder clinical workflows. Additionally, the lack of external validation from diverse centers, absence of standardization in imaging protocols, and limited assessment of model stability across different patient subgroups raise concerns about its broader applicability. Despite the promise of these models, there remains a crucial need for real-world testing in varied clinical environments and continuous validation to ensure relevance with advancements in imaging technologies and methodologies. These limitations underscore that, while the study represents a valuable contribution to the field, more validation, particularly in multi-center settings with larger cohorts, is essential before recommending widespread clinical implementation.

Future directions in research

Xu et al's study has indeed opened new opportunities for advancing the field of radiomics in HCC[1]. To improve the generalizability of predictive models, future research should focus on multicentric studies across diverse populations. This is crucial because genetic, dietary, and environmental factors can influence tumor biology, potentially impacting the radiomics profiles extracted from different groups. Conducting such studies can help fine-tune models for broader applicability. Another key area of development is the integration of radiomics with genetic and molecular markers. By combining imaging data with molecular characteristics, such as gene expression profiles or tumor-infiltrating lymphocytes, researchers can gain deeper insights into the tumor microenvironment, potentially leading to more precise predictive models. This multimodal approach could transform radiomics into an even more powerful tool for personalized medicine[19]. Real-time validation in clinical trials is another necessary step to assess the utility of these models. Prospective trials can evaluate how well these models function in clinical settings and refine them with real-world data, enhancing their robustness. Additionally, a cost-effectiveness analysis is needed to justify the widespread adoption of radiomics in clinical practice. This would involve evaluating the costs associated with software, training, and workflow adjustments[20]. Finally, ethical considerations and data security must not be overlooked. With the increasing reliance on data-centric approaches, ensuring patient consent, data privacy, and security is paramount. Future research must address these issues to build trust and ensure the safe use of patient data in radiomics-driven healthcare[21].

CONCLUSION

The pioneering work by Xu et al[1] marks a significant milestone in HCC management through radiomics-based MVI prediction. The successful translation of these research findings into clinical practice demands robust interdisciplinary collaboration among radiologists, oncologists, surgeons, data scientists, and healthcare informaticians. Such collaboration is essential for establishing standardized protocols, developing comprehensive training programs, integrating radiomics tools into existing clinical workflows, and ensuring regular validation of model performance through multicenter studies. Healthcare institutions should establish dedicated multidisciplinary teams to oversee the implementation while effectively balancing innovation with practical feasibility. Through these coordinated efforts and systematic implementation strategies, combined with continued technological advancement, the future success of radiomics in clinical practice can be realized, ensuring that these advances benefit patients while remaining feasible in real-world clinical settings.