Applying Explainable Machine Learning Models for Detection of Breast Cancer Lymph Node Metastasis in Patients...

Area of Science:

• Oncology
• Medical Informatics
• Computational Biology

Background:

• Increasing use of neoadjuvant systemic therapy (NST) for breast cancer necessitates improved axillary lymph node assessment.
• Current radiological methods have limitations in accurately determining lymph node status, leading to potential misdiagnosis in up to 30% of patients.
• There is a critical need for supplementary methods to enhance lymph node status evaluation.

Purpose of the Study:

• To apply and evaluate machine learning (ML) models for predicting lymph node metastasis.
• To focus specifically on breast cancer patients eligible for NST.
• To assess the utility of clinicopathological data in ML-based lymph node status prediction.

Main Methods:

• Utilized a dataset of 8381 breast cancer patients, identifying 719 eligible for NST.
• Applied various ML models, including random forest and XGBoost, to predict lymph node metastasis.
• Employed Shapley values for model explainability to identify key predictive features.

Main Results:

• In the NST-eligible group, random forest achieved the highest performance with an AUC of 0.793.
• XGBoost demonstrated the best performance in the total study population (AUC: 0.762).
• Shapley value analysis highlighted tumor size, Ki-67, and patient age as the most significant predictors.

Conclusions:

• Tree-based ML models exhibit strong performance in assessing lymph node status.
• These models can enhance disease staging accuracy, leading to optimized treatment selection.
• The findings are particularly relevant for NST patients where accurate lymph node assessment is crucial.