Explainable and uncertainty-aware ensemble framework with causal analysis for breast cancer detection

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Informatics

Background:

• Breast cancer is a leading cause of cancer mortality globally, characterized by aggressive growth and metastasis.
• Current machine learning models for breast cancer diagnosis often lack robust uncertainty handling and clear explainability.
• Addressing these limitations is crucial for improving diagnostic accuracy and clinical trust.

Purpose of the Study:

• To develop an integrated framework for breast cancer prediction that incorporates uncertainty-aware ensemble learning and causal feature analysis.
• To enhance the interpretability and trustworthiness of machine learning models in clinical decision-making.
• To provide clinicians with clear confidence levels for predictions, reducing diagnostic errors.

Main Methods:

• Utilized an ensemble of Light Gradient Boosting Machine (LightGBM), random forest, and gradient boosting classifiers with integrated uncertainty estimation.
• Employed causal analysis to identify potential clinical confounders.
• Integrated multimodal explainability techniques including SHAP (Shapley Additive Explanations), permutation importance, and feature attribution.

Main Results:

• Achieved high performance on two public datasets, with AUCs up to 0.99 and accuracies up to 0.98%.
• Demonstrated 100% precision for high-confidence predictions on one dataset, with no false positives.
• Causal analysis identified key confounders such as lymph node involvement, tumor size, and metastasis; fairness tests indicated balanced performance across demographic groups.

Conclusions:

• The proposed framework effectively combines uncertainty estimation and causal interpretability for accurate and trustworthy breast cancer prediction.
• The model offers clinicians transparent decision support with explicit confidence levels, enhancing reliability in clinical settings.
• This approach has the potential to significantly reduce diagnostic errors and improve patient outcomes.