Predicting Lymph Node Metastasis in Rectal Cancer: Development and Validation of a Machine Learning Model Using...

Area of Science:

• Oncology
• Medical Informatics
• Radiology

Background:

• Rectal cancer (RC) poses a significant health burden, with lymph node metastasis (LNM) critically impacting patient outcomes.
• Traditional diagnostic methods for LNM in RC have limitations, driving the need for advanced predictive tools.
• Clinical data integration offers a promising avenue for developing accurate LNM prediction models.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting LNM risk in rectal cancer patients.
• To assess the performance of eleven different ML algorithms using clinical and imaging data.
• To identify key predictors of LNM in rectal cancer.

Main Methods:

• Retrospective analysis of 2454 rectal cancer patients from the SEER database and an external hospital cohort.
• Integration of computed tomographic (CT) scan-derived lymph node features with clinicopathological data.
• Development and evaluation of eleven ML models, including XGBoost, Random Forest, and Gradient Boosting, using AUC, calibration, and decision curve analysis.

Main Results:

• The XGBoost (XBG) model showed superior cross-cohort stability and minimal calibration error.
• Internal validation AUCs ranged from 0.859 to 0.964, with Random Forest and Extremely Randomized Trees performing highest.
• External validation AUCs ranged from 0.735 to 0.838, with Gradient Boosting and XGBoost showing the best performance.

Conclusions:

• Eleven ML models were successfully developed and validated for LNM risk prediction in RC.
• The XGBoost model emerged as optimal, demonstrating strong predictive power across internal and external validation sets.
• Integrating CT scan data with clinicopathological findings via ML holds significant potential for improving early diagnosis and personalized treatment of RC.