Predicting metastasis in gastric cancer patients: machine learning-based approaches

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Gastric cancer (GC) is a leading cause of cancer mortality globally, with a poor 5-year survival rate below 40%.
• Predicting metastasis is crucial for effective treatment planning and improving patient outcomes in GC.
• Existing predictive methods may not fully leverage the potential of machine learning with comprehensive patient data.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting metastasis status in gastric cancer patients.
• To compare the performance of various ML classifiers, including Naive Bayes, Random Forest, Support Vector Machine, Neural Network, Decision Tree, and Logistic Regression.
• To identify the most effective ML algorithms for predicting GC metastasis based on demographic and clinical variables.

Main Methods:

• Utilized data from 733 gastric cancer patients, divided into training (80%) and testing (20%) sets.
• Applied six ML algorithms (NB, RF, SVM, NN, RT, LR) with 5-fold cross-validation to predict metastasis.
• Assessed model performance using metrics such as F1 score, precision, sensitivity, specificity, ROC AUC, and PR-AUC.

Main Results:

• Out of 733 patients, 262 (36%) exhibited metastasis.
• Support Vector Machine (SVM) and Neural Network (NN) models demonstrated superior performance.
• SVM achieved high AUC (0.85) and sensitivity (0.92) on the test set, while NN showed the highest training AUC (0.98) and testing AUC (0.86).
• Random Forest (RF) was the third most effective model, identifying tumor size and age as key predictors.

Conclusions:

• Machine learning approaches effectively predict gastric cancer metastasis using demographic and clinical characteristics.
• SVM and NN are identified as the most promising algorithms for metastasis prediction in GC.
• These findings can inform the development of advanced diagnostic tools for gastric cancer management.