نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Precise Seizure Onset Detection Using EMD-Based Hybrid Deep-Classical Models on Multichannel EEG

Subject:

Artificial Intelligence and Brain Mapping

Presenter Author:
Name:	Amir hossein Abbasi

Author:
Name 1:	Amir hossein Abbasi - Corresponding - Presenter
Affiliation 1:	Master’s Student in Biomedical Engineering ,School of Advanced Technologies in Medicine, Isfahan University of Medical Sciences, Isfahan, Iran. Email: abbasi_a7852@yahoo.com

Name 2:	Safa Lotfi Gharaei
Affiliation 2:	Master's Student in Cognitive Psychology, Department of Psychology, Ferdowsi University, Mashhad, Iran. Email: safa.lotfigharaei@mail.um.ac.ir

Name 3:
Affiliation 3:

Name 4:
Affiliation 4:

Name 5:
Affiliation 5:

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

Epilepsy is a chronic neurological disorder characterized by recurrent seizures caused by abnormal neuronal discharges. Accurate detection of seizure onset in electroencephalography (EEG) recordings plays a critical role in clinical diagnosis, therapeutic decision-making, and long-term patient monitoring. Despite recent advances in automated seizure detection using machine learning, reliable identification of seizure onset remains challenging due to inter-subject variability and the complex, nonstationary nature of multichannel EEG signals.

Methods:

This study utilizes multichannel scalp EEG recordings from six patients with focal epilepsy acquired at the American University of Beirut Medical Center. The dataset consists of 19 EEG channels arranged according to the international 10–20 system, sampled at 500 Hz, and categorized into four classes: normal activity, complex partial seizure (CPS), electrographic seizure, and video-detected seizure without visible EEG changes. After bandpass filtering (0.5–40 Hz), signals were segmented into two-second epochs. Empirical Mode Decomposition (EMD) was applied to each epoch to extract intrinsic mode functions (IMFs), capturing the nonstationary characteristics of EEG signals. A set of temporal, statistical, and nonlinear features was extracted from the IMFs and reduced using Principal Component Analysis (PCA). Both classical classifiers (Support Vector Machine and Decision Tree) and hybrid models integrating a Convolutional Neural Network (CNN) with classical classifiers were evaluated using five-fold cross-validation and an independent test set.

Results:

The experimental results demonstrate that the proposed EMD-based feature extraction framework provides highly discriminative representations for seizure onset classification. The hybrid CNN-SVM model achieved improved training accuracy (84.91% ± 2.00%) compared to the standalone SVM (82.81% ± 1.92%), indicating that CNN-based feature refinement enhances classical classification performance. Among all evaluated models, the Decision Tree classifier achieved the highest test accuracy of 97.18%, exhibiting strong generalization capability across unseen data. These findings highlight the effectiveness of feature-driven models as well as the complementary role of hybrid architectures.

Conclusion:

This work demonstrates that combining EMD-based feature extraction with hybrid deep–classical machine learning models enables accurate and reliable seizure onset detection from scalp EEG recordings. While the superior generalization performance of the Decision Tree emphasizes the importance of well-engineered features, the improved training performance of the hybrid CNN-SVM validates the proposed hybrid learning strategy. The presented framework establishes a robust foundation for future developments, including architectural optimization, evaluation on larger multi-center datasets, and enhancement of cross-subject generalization toward clinically deployable automated epilepsy monitoring systems.

Keywords:

Seizure, Electroencephalography, Empirical Mode Decomposition, Hybrid Deep–Classical Learning, Multichannel Signal Analysis