Accuracy of Medical Image-Based Deep Learning for Detecting Microvascular Invasion in Hepatocellular Carcinoma:...

Area of Science:

• Hepatocellular carcinoma (HCC) research
• Medical imaging analysis
• Artificial intelligence in oncology

Background:

• Hepatocellular carcinoma (HCC) is a major cause of cancer mortality.
• Microvascular invasion (MVI) is a key predictor of poor prognosis and recurrence in HCC patients.
• Deep learning (DL) shows potential for diagnosing MVI from medical images.

Purpose of the Study:

• To systematically evaluate the diagnostic performance of DL models for preoperative MVI prediction in HCC using medical images.
• To assess the impact of imaging modalities and validation strategies on DL model performance and generalizability.

Main Methods:

• A meta-analysis of studies using imaging-based DL for MVI detection in HCC.
• Searched PubMed, Cochrane Library, Embase, and Web of Science.
• Bivariate mixed-effects meta-analysis to calculate pooled sensitivity, specificity, and SROC.
• Quality assessment using the QUADAS-2 tool.
• Subgroup analyses by imaging modality and validation method.

Main Results:

• Included 52 studies with 19,531 HCC patients.
• Overall DL model performance: sensitivity 0.80, specificity 0.82, SROC 0.88.
• Contrast-enhanced CT models showed high performance (SROC 0.90).
• Pathological section DL models achieved the highest performance (SROC 0.92).
• Model performance was less consistent in external validation (SROC 0.85) vs. internal validation (SROC 0.90).

Conclusions:

• Imaging-based DL models demonstrate significant potential for MVI prediction in HCC.
• Contrast-enhanced CT is a promising noninvasive method.
• Rigorous external validation is essential to avoid overestimating model efficacy.
• Future research should focus on prospective, multicenter studies with standardized reporting and integrated algorithms.