A comprehensive explainable AI approach for enhancing transparency and interpretability in stroke prediction.

Area of Science:

• Medical Informatics
• Computational Medicine
• Artificial Intelligence in Healthcare

Background:

• Stroke is a leading cause of death and disability, particularly in older adults.
• Early diagnosis and treatment are crucial for reducing mortality and severe cerebral disability.
• Healthcare professionals can benefit from advanced tools for more effective and immediate stroke management.

Purpose of the Study:

• To investigate the application of machine learning (ML) classifiers and explainable artificial intelligence (XAI) for stroke prediction.
• To identify key features indicative of stroke risk using various feature selection methodologies.
• To provide preliminary insights for developing AI-driven tools to assist clinicians in stroke patient management.

Main Methods:

• Utilized six different ML classifiers trained on stroke patient datasets.
• Employed six feature selection methodologies to extract critical features from the data.
• Applied XAI techniques including Shapley Additive Values (SHAP), ELI5, and Local Interpretable Model-agnostic Explanations (LIME) for model interpretability.

Main Results:

• Successfully applied ML classifiers and XAI methods to predict stroke.
• Identified essential features contributing to stroke prediction through feature selection.
• Demonstrated the potential of XAI in understanding the factors influencing stroke prediction models.

Conclusions:

• AI and ML show promise in enhancing early stroke detection and management.
• XAI methods offer valuable insights into the decision-making processes of ML models for stroke prediction.
• Further clinical validation and real-world testing are necessary to translate these findings into practical healthcare tools.