Radiomics-based models for the preoperative prediction of high-risk pathological features in gastric cancer: a systematic review and meta-analysis

Gastric cancer remains a major global health burden and one of the leading causes of cancer-related morbidity and mortality worldwide. Accurate preoperative identification of high-risk pathological features, including perineural invasion (PNI), lymph node metastasis (LNM), lymphovascular invasion (LVI), and serosal invasion, is essential for individualized treatment planning. Radiomics enables the extraction of high-throughput quantitative features from preoperative medical images, which can be integrated with machine learning or deep learning algorithms to develop predictive models for noninvasive risk stratification of high-risk pathological features in gastric cancer. This systematic review and meta-analysis was reported in accordance with PRISMA-DTA. PubMed, Web of Science, Cochrane Library, and Embase were comprehensively reviewed, including studies until May 11, 2026, to evaluate radiomics models for the preoperative prediction of PNI, LNM, LVI, and serosal invasion in gastric cancer. Study quality was assessed using a modified Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool and the Radiomics Quality Score (RQS). Predictive performance was pooled using a bivariate random-effects model. A three-level random-effects model was further applied to account for multiple effect sizes within the same study. A total of 40 studies involving 12,769 patients were included. In the validation cohorts, combined models yielded AUCs of 0.85 (95% CI, 0.82–0.88) for PNI, 0.85 (95% CI, 0.82–0.88) for LNM, and 0.86 (95% CI, 0.82–0.89) for LVI. Seven studies evaluated serosal invasion, with five contributing to the combined-model analysis, which yielded an AUC of 0.90 (95% CI, 0.87–0.93); the 95% prediction interval for specificity was wide (0.44–0.97). Combined models showed predictive potential for preoperative assessment of high-risk pathological features in gastric cancer; however, the evidence was derived predominantly from internal validation. Prospective multicenter external validation is required before clinical implementation. PROSPERO CRD420251071990.