Biomarker-based and interpretable machine learning framework for predicting pathological stage in gastric cancer: A...

Area of Science:

• Oncology
• Bioinformatics
• Medical Diagnostics

Background:

• Accurate preoperative staging of gastric cancer (GC) is crucial for effective treatment.
• Current noninvasive methods for distinguishing early from advanced GC are limited.

Purpose of the Study:

• To develop and validate a machine learning model for noninvasive GC staging using routine laboratory parameters.
• To identify key predictive features for improved diagnostic accuracy.

Main Methods:

• Retrospective analysis of 434 GC patients' preoperative laboratory data.
• Development of 11 supervised machine learning algorithms, with CatBoost selected for its superior performance.
• Feature engineering included inflammatory, metabolic, and tumor-related ratio profiles; SHAP analysis identified key predictors.

Main Results:

• The CatBoost model incorporating ratio features achieved a high Area Under the Curve (AUC) of 0.9499.
• SHAP analysis identified uric acid (UA) and APTT as significant predictors.
• The model demonstrated excellent discrimination, interpretability, and robustness through cross-validation.

Conclusions:

• A robust and interpretable machine learning model for GC staging was developed using routine blood tests.
• The model offers a practical tool for personalized risk stratification and treatment planning in gastric cancer.