Curriculum Learning with Sampling Scheduler for Imbalanced EEG-Based Seizure Detection

Epileptic seizures pose serious health risks and significantly affect the quality of life for individuals with epilepsy, emphasizing the importance of accurate and timely detection. Despite advancements in electroencephalography (EEG) based seizure detection, class imbalance between seizure and non-...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:The ... International Winter Conference on Brain-Computer Interface S. 1 - 5
Hauptverfasser: Choi, WooHyeok, Kim, Jun-Mo, Nam, Hyeonyeong, Kam, Tae-Eui
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 24.02.2025
Schlagworte:
Accuracy
Brain modeling
Class Imbalance
Curriculum Learning
Data models
Dynamic scheduling
Electroen-cephalogram
Electroencephalography
Epilepsy
Epileptic Seizure
Hospitals
Machine learning
Sampling Scheduler
Scalp
Seizure Detection
Training
ISSN:2572-7672
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Epileptic seizures pose serious health risks and significantly affect the quality of life for individuals with epilepsy, emphasizing the importance of accurate and timely detection. Despite advancements in electroencephalography (EEG) based seizure detection, class imbalance between seizure and non-seizure classes remains a major challenge, often leading to biased machine learning models and reduced generalizability. To address this, we propose a sampling scheduler approach inspired by dynamic curriculum learning. Unlike conventional resampling techniques, our method gradually adjusts the sampling ratio between seizure and non-seizure samples during training, enabling the model to effectively learn from imbalanced data. Experiments on the children's hospital boston and the massachusetts institute of technology scalp EEG (CHB-MIT) dataset demonstrate that our approach outperforms conventional methods, achieving superior results in balanced accuracy, specificity, area under th curve, and geometirc mean. These findings highlight the potential of the sampling scheduler to address the imbalance problem and enhance EEG-based seizure detection.
ISSN:2572-7672
DOI:10.1109/BCI65088.2025.10931459