Machine Learning-Based Diagnostic Models for Early Gastric Cancer Using Clinical Laboratory Indicators

Area of Science:

• Oncology
• Bioinformatics
• Medical Diagnostics

Background:

• Gastric cancer involves complex pathological processes with numerous clinical indicator abnormalities.
• Machine learning offers advanced capabilities for analyzing extensive variables in disease prediction and diagnosis.

Purpose of the Study:

• To develop and evaluate machine learning models for accurate gastric cancer diagnosis.
• To identify key indicators contributing to gastric cancer prediction.

Main Methods:

• Collected clinical data from gastric cancer patients (2016-2023).
• Applied five machine learning algorithms: XGBoost, RF, SVM-RFE, LGBM, and rpart.
• Evaluated model performance using AUROC, F1-score, sensitivity, and specificity.

Main Results:

• XGBoost achieved the highest diagnostic performance (AUC=0.9909) when combining blood and pathological data.
• Key diagnostic indicators included Glutathione reductase (GR), CA724, RBC, CA242, and ALB.
• Tumor size was identified as an independent risk factor for early gastric cancer.

Conclusions:

• Machine learning models integrating blood and pathological data enhance gastric cancer risk prediction accuracy.
• The XGBoost model exhibits excellent diagnostic performance, supporting preclinical implementation.
• This study provides evidence for the clinical utility of machine learning in gastric cancer diagnostics.