Using Explainable Artificial Intelligence to Obtain Efficient Seizure-Detection Models Based on...

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Epilepsy affects 50 million globally, causing seizures with diverse manifestations.
• Seizures significantly impact quality of life, leading to social isolation and distress.
• Current detection methods often rely on complex machine learning or deep learning on EEG signals.

Purpose of the Study:

• To develop a simplified, explainable artificial intelligence (XAI) methodology for epileptic seizure detection.
• To reduce the number of features and EEG channels required for accurate seizure detection.
• To validate the effectiveness of simpler models for seizure detection without deep learning.

Main Methods:

• Utilized Explainable Artificial Intelligence (XAI) for epileptic seizure detection.
• Implemented a feature and channel reduction strategy for simpler classifiers.
• Performed temporal domain analysis on EEG signals within a 1-second time window.

Main Results:

• Achieved performance metrics exceeding 95% in accuracy, precision, recall, and F1-score.
• Successfully detected epileptic seizures using only six features and five EEG channels.
• Demonstrated robust generalization across a diverse patient cohort.

Conclusions:

• Feature reduction in simpler models is adequate for effective epileptic seizure detection.
• Strategic selection of electrodes and reduced attributes can support effective mobile seizure detection applications.
• The proposed XAI methodology offers a promising, less complex alternative for seizure detection.