Establishment of a machine learning model for predicting splenic hilar lymph node metastasis

Area of Science:

• Oncology
• Medical Statistics
• Surgical Oncology

Background:

• Upper gastrointestinal cancer (UGC) can metastasize to splenic hilum lymph nodes (SHLN), posing surgical challenges and risks.
• Current predictive models for SHLN metastasis often lack sufficient detail, providing only point estimates.
• Accurate prediction of SHLN metastasis is crucial for surgical planning and patient management.

Purpose of the Study:

• To develop and evaluate a novel Bayesian logistic regression model, Bayes-SHLNM, for predicting SHLN metastasis in UGC.
• To compare the performance of Bayes-SHLNM against a traditional frequentist logistic regression (FLR) model.
• To visualize the uncertainty associated with metastasis predictions using posterior probability distributions (PPD).

Main Methods:

• Development of a Bayesian logistic regression model (Bayes-SHLNM) for SHLN metastasis prediction.
• Comparison of Bayes-SHLNM performance with a frequentist logistic regression (FLR) benchmark model.
• Individualized visualization of posterior probability distributions (PPD) to represent prediction uncertainty.

Main Results:

• The Bayes-SHLNM model demonstrated performance equivalent to the FLR model in predicting SHLN metastasis.
• The PPD for each case was successfully visualized, effectively communicating prediction uncertainty.
• The Bayesian approach provides a more comprehensive understanding of risk compared to point estimates.

Conclusions:

• The Bayes-SHLNM model is a viable tool for predicting SHLN metastasis in upper gastrointestinal cancer.
• Visualizing PPD enhances clinical decision-making by quantifying prediction uncertainty.
• Bayes-SHLNM shows potential as a clinical decision support system, particularly in complex cases with high uncertainty.