Objective:

To synthesize existing evidence to evaluate the overall diagnostic accuracy and sources of heterogeneity in image-based AI models for predicting VETC status and its prognostic value in HCC.

Approach:

• Systematic Review and Meta-Analysis: Adhered to PRISMA-DTA guidelines, CHARMS checklists, and was prospectively registered on PROSPERO.
• Literature Search: Conducted across four major databases: PubMed, Embase, Web of Science, and Cochrane Library, with updates for reproducibility and a specified time frame.

Key Findings:

• Vessels encapsulating tumor clusters (VETCs) are an independent poor prognostic factor in HCC.
• AI technologies, particularly deep learning and radiomics, show potential for noninvasive prediction of VETC status.
• Existing studies exhibit variability in model performance and lack methodological standardization.

Interpretation:

The review addresses gaps in existing literature by pooling diagnostic accuracy across multiple imaging modalities and integrating risk-of-bias appraisal.

Limitations:

• Variability in model performance across studies impacts the reliability of findings.
• Small sample sizes in existing research limit the generalizability of results.
• Lack of methodological standardization in AI applications affects comparability.

Conclusion:

This review provides evidence for preoperative decision-making regarding VETC in HCC.