Predicting gastric cancer survival using machine learning: A systematic review.

Area of Science:

• Oncology
• Medical Informatics
• Computational Biology

Background:

• Gastric cancer (GC) survival prediction is challenging.
• Machine learning (ML) offers potential but faces interpretability and data limitations.

Purpose of the Study:

• Evaluate ML applications for GC survival prediction.
• Identify key limitations in current ML methods for GC.

Main Methods:

• Systematic literature review of 16 studies (post-2019) from PubMed/Web of Science.
• Analysis of ML model types (Deep Learning, Random Forests, SVMs, Ensemble) and dataset sizes.
• Inclusion of studies with external validation.

Main Results:

• ML models achieved AUCs of 0.669-0.980 (overall survival), 0.920-0.960 (cancer-specific), and 0.710-0.856 (disease-free).
• Demonstrated potential for personalized treatment planning and risk stratification in GC patients.

Conclusions:

• ML models show significant promise for GC survival prediction.
• Challenges include retrospective data reliance and lack of interpretability.
• Recommendations include prospective trials and multidimensional data integration.