Comparative Effectiveness of Artificial Intelligence Versus Conventional Methods for Detecting Peritoneal Metastasis in Colorectal Cancer: A Systematic Review.

Abstract

Colorectal cancer represents a major global malignancy and a leading cause of cancer-related death. Peritoneal metastasis occurs in a significant proportion of colorectal cancer patients and is associated with markedly worse prognosis compared to other metastatic sites, with a limited median overall survival. Early detection remains challenging due to the limited sensitivity of conventional imaging techniques, with computed tomography exhibiting poor detection rates for small lesions and necessitating invasive diagnostic procedures for accurate diagnosis. The limitations of traditional diagnostic modalities have driven a growing interest in artificial intelligence applications to advance the early, non-invasive detection of peritoneal metastasis. This study aimed to systematically assess whether artificial intelligence and machine learning approaches enhance the accuracy and efficiency of detecting peritoneal metastasis and predicting tumor spread patterns compared to conventional imaging and clinical assessment methods in patients with colorectal cancer. A systematic review was conducted in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, searching the PubMed, Web of Science, Cochrane, Embase, and Scopus databases for studies published between 2015 and 2025. The search strategy included comprehensive terminology related to artificial intelligence and machine learning, combined with terms related to peritoneal metastasis. Two independent reviewers assessed study quality using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool for diagnostic accuracy studies and the modified Radiomics Quality Score for artificial intelligence (AI) and radiomics studies, with disagreements resolved through consensus discussions. From multiple countries, 22 studies were included with a total population of over 40,000 patients. AI applications consistently outperformed traditional methods across all modalities. While conventional approaches showed moderate performance with C-indices of 0.73-0.85 and CT imaging missed 89% of small lesions, AI-assisted systems demonstrated superior results as follows: cytological detection achieved over 95% accuracy and 99% specificity; radiomics models reached AUCs up to 0.941; circulating tumor DNA integration provided 8.5-fold increased risk identification; and computer-assisted staging laparoscopy improved surgical diagnostic accuracy from 52% to 79% compared to human assessment alone. AI technologies demonstrate promising advantages for peritoneal metastasis detection, offering enhanced diagnostic accuracy, objective assessments, faster analysis, and improved clinical decision-making, particularly through human-AI collaboration. However, most studies lack external validation across diverse populations and real-world settings, while current implementations face significant workflow challenges. Before clinical adoption, future research must prioritize large-scale prospective validation studies, external validation across diverse populations, and comprehensive cost-effectiveness analyses to ensure safe and effective integration into clinical practice.