Epileptic seizure detection using EEG signals and extreme gradient boosting

Area of Science:

• * Computational neuroscience and biomedical signal processing.
• * Application of artificial intelligence in clinical diagnostics.

Background:

• * Automated seizure detection from electroencephalograms (EEG) using machine learning faces performance challenges with existing datasets.
• * The Temple University Hospital EEG Seizure Corpus (TUSZ) is a complex dataset for evaluating seizure detection algorithms.
• * Current performance levels do not meet expectations for clinical utility.

Purpose of the Study:

• * To investigate the impact of increased data volume on the performance of automated seizure detection.
• * To evaluate machine learning model efficacy with enhanced training datasets.
• * To identify key features and EEG channels crucial for accurate seizure classification.

Main Methods:

• * Utilized the Temple University Hospital EEG Seizure Corpus (TUSZ) with a focus on larger, recent versions.
• * Employed two data augmentation strategies: standard partitioning and a leave-one-out approach for training set expansion.
• * Implemented XGBoost, a gradient boosting classifier, for its efficiency in handling large datasets.

Main Results:

• * Performance achieved is comparable to state-of-the-art deep learning models in the literature.
• * Generalized seizures demonstrated significantly higher prediction accuracy compared to focal seizures.
• * Identified specific EEG channels and signal features as critical for distinguishing seizure activity from background brain activity.

Conclusions:

• * Augmenting training data significantly improves automated seizure detection performance.
• * Machine learning models, like XGBoost, can achieve competitive results with sufficient data.
• * Seizure type and specific EEG characteristics are important factors influencing detection accuracy.