Explainable Artificial Intelligence Model for Stroke Prediction Using EEG Signal.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Technology

Background:

• Artificial Intelligence (AI) models are increasingly used in healthcare for disease diagnosis but often function as "black boxes," lacking transparency.
• Explainable AI (XAI) addresses this by providing insights into AI decision-making processes.
• Electroencephalography (EEG) shows promise for predicting ischemic stroke, aiding in diagnosis, prognosis, and treatment.

Purpose of the Study:

• To develop and validate Machine Learning (ML) models for classifying acute ischemic stroke patients and healthy controls using EEG data.
• To apply XAI techniques (Eli5 and LIME) to interpret ML model behavior and identify key predictive features.
• To enhance the explainability of AI-driven stroke prediction for clinical application.

Main Methods:

• Collected EEG data from 48 acute ischemic stroke patients and 75 healthy adults during active states (walking, working, reading).
• Utilized EEG recordings from frontal, central, temporal, and occipital cortical electrodes within three months of stroke onset.
• Employed Adaptive Gradient Boosting ML models and XAI tools (Eli5, LIME) for classification and feature interpretation.

Main Results:

• Adaptive Gradient Boosting models achieved approximately 80% accuracy in distinguishing between stroke and control groups.
• XAI tools identified spectral delta and theta features as significant local contributors to stroke prediction.
• The study successfully explained the ML model's predictions, highlighting specific EEG patterns.

Conclusions:

• XAI models can effectively interpret ML-based stroke prediction using EEG data.
• The identified spectral delta and theta features are crucial for acute ischemic stroke prediction.
• This research facilitates more transparent and reliable AI-assisted diagnostics in stroke care, supporting treatment and recovery.