A Cost-Effective Model for Predicting Recurrent Gastric Cancer Using Clinical Features

Area of Science:

• Oncology
• Biostatistics
• Artificial Intelligence

Background:

• Gastric cancer recurrence poses a significant challenge for survivors.
• Identifying reliable clinical biomarkers for recurrence is crucial for improved patient management.

Purpose of the Study:

• To utilize artificial intelligence (AI) techniques to identify clinical biomarkers for predicting recurrence in gastric cancer survivors.
• To benchmark various AI algorithms for their effectiveness in this prediction task.

Main Methods:

• Employed Random Forest, MLP, C4.5, AdaBoost, and Bagging algorithms on a dataset of 2476 gastric cancer survivors.
• Utilized Synthetic Minority Oversampling Technique (SMOTE) for imbalanced data, cost-sensitive learning for risk assessment, and SHapley Additive exPlanations (SHAPs) for feature importance.

Main Results:

• The proposed Random Forest model achieved high performance with 87.9% accuracy, 90.5% recall, 86% precision, and 88.2% F1-score on a balanced dataset.
• Identified the top five clinical features influencing recurrence prediction: stage, lymph node involvement, Helicobacter pylori infection, BMI, and gender.

Conclusions:

• AI models, particularly Random Forest, can effectively identify and rank risk factors for recurrent gastric cancer.
• The identified clinical features are significant predictors and can assist physicians in screening high-risk gastric cancer survivors.