نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Design of a Drowsiness Detection System Based on Graph Theory-Based Features of EEG Signals

Subject:

Artificial Intelligence and Brain Mapping

Presenter Author:
Name:	Sarina Sheykhkazemi

Author:
Name 1:	Sarina Sheykhkazemi - Presenter
Affiliation 1:	Department of Biomedical Engineering Amirkabir University of Technology Tehran, Iran

Name 2:	Golnaz Baghdadi - Corresponding
Affiliation 2:	Department of Biomedical Engineering Amirkabir University of Technology Tehran, Iran

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 :

Driver drowsiness is a major safety concern because reduced vigilance can impair attention, reaction time, and decision-making during driving. Electroencephalography (EEG) provides a direct measure of brain activity and has been widely investigated for detecting changes in vigilance. However, conventional EEG features mainly characterize individual channels and may not fully capture interactions between distributed brain regions. Functional connectivity and graph theory provide an alternative approach by representing the brain as an interconnected network and quantifying its topological organization. This study aims to systematically compare Phase Lag Index (PLI)- and Phase Locking Value (PLV)-based functional connectivity using graph-theoretical features for distinguishing alert and drowsy states within an interpretable machine-learning framework.

Methods:

A publicly available multichannel EEG dataset recorded during a sustained-attention virtual-reality driving task was used. EEG recordings from 11 participants were analyzed, with the corresponding recordings categorized into alert and drowsy states based on behavioral response measures. EEG signals were preprocessed using MATLAB and EEGLAB and decomposed into seven frequency sub-bands: low theta (4–6 Hz), high theta (6–8 Hz), low alpha (8–10 Hz), high alpha (10–13 Hz), low beta (13–18 Hz), mid beta (18–25 Hz), and high beta (25–30 Hz). For each frequency band, functional connectivity matrices were independently constructed using PLI and PLV. Five weighted graph-theoretical measures, including node strength, clustering coefficient, characteristic path length, global efficiency, and local efficiency, were extracted from each connectivity network. Correlation-based feature reduction was performed using a threshold of 0.95 to reduce redundancy. Seven conventional machine-learning classifiers were evaluated using Leave-One-Out Cross-Validation (LOO-CV), including Linear Discriminant Analysis, Logistic Regression, Decision Tree, Naive Bayes, K-Nearest Neighbors, Random Forest, and Support Vector Machine.

Results:

Among the PLI-based models, the linear Support Vector Machine achieved the highest classification performance, with with a Leave-One-Subject-Out cross-validation accuracy of 81.8%. In contrast, the best-performing PLV-based model, obtained using Linear Discriminant Analysis, achieved an accuracy, precision, and recall of approximately 45.5%, with an F1-score of 0.45. Feature importance analysis of the best-performing PLI-SVM model indicated that connectivity strength in the low-alpha, high-alpha, and low-theta bands contributed most strongly to classification. Characteristic path length in the mid-beta, high-beta, and low-theta bands also showed substantial contributions. These findings suggest that PLI-derived graph features captured informative network-level characteristics associated with changes in vigilance.

Conclusion:

The findings demonstrate the potential of combining PLI-based functional connectivity, graph-theoretical analysis, and conventional machine learning for EEG-based driver drowsiness detection. The importance of alpha- and theta-band features further indicates that changes in functional brain-network organization may provide useful information for distinguishing alert and drowsy states. Despite the limited number of participants, the proposed framework offers an interpretable approach for investigating network-level EEG characteristics associated with drowsiness.

Keywords:

Driver drowsiness detection, electroencephalography, functional connectivity, graph theory, phase lag index, machine learning.