Interpretable Machine Learning for Predicting Neoadjuvant Chemotherapy Response in Breast Cancer Using the Baseline...

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Pathological response to neoadjuvant chemotherapy (NAC) is a key prognostic indicator in breast cancer (BC).
• Existing models often lack sufficient predictive power due to limited features.
• Accurate prediction of treatment response is crucial for optimizing patient management.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting pathological complete response (pCR) to NAC.
• To utilize baseline clinical and pathological features for enhanced prediction accuracy.
• To improve treatment strategy optimization for BC patients.

Main Methods:

• Collected data from 303 BC patients treated with NAC.
• Employed LASSO regression for feature selection.
• Developed and compared six ML models (XGBoost, LightGBM, CatBoost, logistic regression, RF, SVM).
• Assessed model performance using AUC, accuracy, precision, recall, F1, and Brier scores.
• Utilized SHAP for model interpretability.

Main Results:

• The CatBoost model achieved the highest predictive performance with an AUC of 0.853 after hyperparameter tuning.
• Twelve features, including stromal tumor-infiltrating lymphocytes (sTILs), were identified as significant predictors.
• sTILs were identified as the most critical predictive feature by SHAP analysis.
• The CatBoost model with sTILs demonstrated an average AUC of 0.83 in cross-validation.

Conclusions:

• An ML-based model can accurately predict pCR in BC patients at baseline.
• This prediction aids in optimizing NAC treatment strategies.
• The interpretable SHAP framework increases clinical trust and understanding of the ML model.