Journal of Biomedical Advancement Scientific Research

Abstract

Background: Accurate preoperative identification of peritoneal metastasis remains a major challenge in patients with gastric cancer (GC). Machine learning (ML) techniques may improve risk prediction by inte grating multiple clinicopathological variables.

Methods: A total of 809 patients diagnosed with GC were included in this study, comprising 712 patients without peritoneal metastasis and 97 patients with metastatic disease. The dataset was randomly divided into training (80%) and testing (20%) cohorts. Six ML algorithms—Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Naïve Bayes (NB)—were developed and compared. Variable importance analyses were performed to identify the most influential predictors, and model performance was evaluated using standard classification metrics.

Results: The most informative factors associated with peritoneal metastasis included lymph node involve ment, number of lymph nodes removed, depth of tumor invasion, lymphatic vessel invasion, and the extent of lymph node dissection. Among systemic inflammatory biomarkers, the platelet-to-lymphocyte ratio (PLR) showed a significant association with metastatic status (p = 0.018), whereas the neutrophil-to-lymphocyte ratio (NLR) did not reach statistical significance (p = 0.121). Of the evaluated algorithms, the RF model achieved the highest predictive accuracy (97%), outperforming SVM and LR.

Conclusions: Machine learning models demonstrate strong potential for predicting peritoneal metastasis in gastric cancer patients. Random Forest provided the most accurate classification performance, sug gesting its utility as a clinical decision-support tool for risk assessment. Further validation using external, multicenter datasets is necessary before implementation in routine practice.