Explainable machine learning model for predicting internal mammary node metastasis in breast cancer: Multi-method...

Area of Science:

• Oncology
• Machine Learning
• Biostatistics

Background:

• Breast cancer diagnosis and staging are critical for treatment.
• Internal mammary lymph node metastasis (IMNM) is an important prognostic factor.
• Accurate baseline assessment of IMNM is crucial for personalized treatment strategies.

Purpose of the Study:

• To develop and validate an explainable machine learning model for predicting baseline IMNM in breast cancer patients.
• To identify key clinical features associated with IMNM.
• To provide a practical tool for early risk stratification.

Main Methods:

• Utilized three cohorts for model development and validation (derivation, internal testing, SEER).
• Employed various machine learning techniques including LASSO, Boruta, and SVM for feature selection and model construction.
• Applied Shapley Additive Explanations (SHAP) for model interpretability.

Main Results:

• A predictive model using six clinical features (N stage, size, stage, classification, grade, location) demonstrated robust performance (AUCs ranging from 0.806 to 0.864).
• High-risk patients identified by the model showed significantly worse outcomes in disease-free survival (DFS) and overall survival (OS).
• An online prediction tool with SHAP-based explanations was created.

Conclusions:

• The developed machine learning model is validated and explainable.
• It serves as a practical tool for early risk stratification of IMNM.
• Aids clinicians in selecting appropriate baseline imaging and planning adjuvant treatment.