Interpretable Machine Learning Models for Predicting Lateral Pelvic Lymph Node Metastasis in Rectal Cancer: A Chinese...

Area of Science:

• Oncology
• Medical Informatics
• Radiology

Background:

• Internal iliac and obturator lymph nodes are frequent sites of rectal cancer metastasis.
• Accurate prediction of lymph node metastasis is crucial for effective treatment planning.

Purpose of the Study:

• To develop and interpret a machine learning (ML) model for predicting lymph node metastasis in rectal cancer.
• To identify key clinical features influencing metastasis prediction using Shapley Additive explanations (SHAP).

Main Methods:

• Retrospective collection of clinical data from rectal cancer patients across four Chinese centers.
• Development and evaluation of five ML models, including Random Forest (RF), using AUC, accuracy, and F1 score.
• Application of SHAP analysis to interpret feature importance and model predictions.

Main Results:

• The RF model demonstrated superior performance in both test and external validation sets (AUC up to 0.899, accuracy up to 0.827).
• Key predictors identified by SHAP analysis included short-axis diameter of enlarged lymph nodes, regional lymph node metastasis, and tumor-to-anal verge distance.
• SHAP force plots provided individual patient-level explanations for model predictions.

Conclusions:

• An interpretable ML model accurately predicts lymph node metastasis in rectal cancer using clinical data.
• SHAP analysis enhances the understanding of feature contributions, supporting personalized treatment decisions.