Key points are not available for this paper at this time.
BACKGROUND: Liver metastasis drives poor outcomes in pancreatic cancer, often precluding curative surgery. Traditional imaging misses occult lesions, risking futile resections. AI models integrating radiomics and clinical data promise enhanced preoperative detection, but evidence synthesis is lacking. METHODS: This systematic review and meta-analysis followed PRISMA-DTA guidelines. We searched PubMed and Web of Science (inception to May 2025) for studies using AI to predict liver metastasis in pancreatic ductal adenocarcinoma (PDAC). Diagnostic metrics were pooled via bivariate random-effects modeling. RESULTS: Of 10 included studies, 6 (9,887 patients) enabled meta-analysis. Pooled AUC was 0.86 (95% CI: 0.83-0.89), sensitivity 0.78 (95% CI: 0.67-0.86), and specificity 0.80 (95% CI: 0.69-0.88). Larger sample sizes correlated with superior performance (P = 0.0471). No publication bias was detected. CONCLUSIONS: AI models offer robust, noninvasive tools for predicting liver metastasis, enabling precise preoperative stratification to optimize surgical decisions and reduce unnecessary procedures. This review systematically summarizes available evidence on AI models predicting liver metastasis and provides practical implications for their clinical application in PDAC.
Wu et al. (Wed,) studied this question.