A Hybrid Algorithm of ML and XAI to Prevent Breast Cancer: A Strategy to Support Decision Making

Area of Science:

• Oncology
• Health Informatics
• Artificial Intelligence

Background:

• The COVID-19 pandemic has exacerbated healthcare challenges, particularly in cancer patient care.
• Breast cancer management, including prevention, diagnosis, and treatment, has been significantly impacted globally.
• Over 20 million breast cancer cases and 10 million deaths were recorded by 2020, highlighting the need for improved strategies.

Purpose of the Study:

• To develop a decision support strategy for healthcare teams using machine learning (ML) and explainability algorithms (XAI).
• To evaluate ML algorithms for classifying cancer patients and to interpret predictive variables using XAI.
• To enhance early detection and personalized patient care in breast cancer management.

Main Methods:

• Evaluation of various ML algorithms for patient classification (cancer vs. non-cancer).
• Integration of an ML methodology with an XAI algorithm (SHAP) for disease prediction and variable interpretation.
• Utilizing the XGBoost algorithm for its predictive capabilities.

Main Results:

• The XGBoost algorithm demonstrated strong predictive performance with an accuracy of 0.813 (train) and 0.81 (test).
• The SHAP algorithm successfully identified significant variables influencing cancer prediction.
• Quantified the impact of these variables on patient health, enabling personalized risk assessment.

Conclusions:

• The proposed ML and XAI strategy offers a robust tool for breast cancer patient classification and prediction.
• Explainability algorithms provide crucial insights into predictive factors, aiding clinical decision-making.
• This approach facilitates early, personalized alerts for patients, potentially improving health outcomes.