Artificial intelligence in predicting lymphovascular invasion in colorectal cancer imaging: a systematic review and meta-analysis

Lymphovascular invasion (LVI) is a critical prognostic factor in colorectal cancer (CRC), strongly associated with poor survival outcomes. Currently, LVI diagnosis relies entirely on postoperative histopathological examination, which is an invasive intervention. This study aims to systematically review and meta-analyze the diagnostic performance of artificial intelligence (AI) models (radiomics and deep learning) for the non-invasive, preoperative prediction of LVI using pre-operative medical imaging. A comprehensive search of PubMed, Embase, Scopus, and Web of Science was performed up to October 3, 2025. Eligible studies included patients with histopathologically confirmed CRC who underwent preoperative imaging (MRI, CT, or PET) analyzed by AI models. Quality assessment was performed using the QUADAS-2 and the RQS 2.0 tools. Pooled sensitivity and specificity were calculated using a random-effects model. A total of 18 retrospective studies encompassing 5,524 patients were included in the systematic review, of which 17 provided sufficient data for quantitative synthesis. All included studies were retrospective and conducted in China, limiting the generalizability of the pooled findings to other populations and healthcare settings. The pooled sensitivity and specificity were 0.77 (95% CI: 0.70–0.82) and 0.81 (95% CI: 0.77–0.85), respectively. The SROC curve yielded a total AUC of 0.84 (pAUC: 0.74), indicating robust performance. At a pre-test probability of 38%, the pooled estimates corresponded to a positive predictive value of 72% and a negative predictive value of 85%. These prevalence-dependent estimates suggest potential utility for preoperative risk stratification but require prospective clinical validation. However, the mean RQS 2.0 score was 24.7 (44%), highlighting important methodological limitations in the current literature. AI-based imaging analysis shows promising diagnostic accuracy for preoperative LVI prediction in CRC and may serve as a complementary tool for risk stratification. However, prospective, multicenter validation in geographically and ethnically diverse populations, together with standardized imaging protocols, is required before broader clinical implementation.