نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Automated Epilepsy Detection from EEG Signals Using Convolutional Neural Networks: A Comparative Study with Classical Machine Learning Methods

Subject:

Advanced Techniques in Signal Acquisition and Image Processing

Presenter Author:
Name:	Yasamin Baghersad

Author:
Name 1:	Yasamin Baghersad - Presenter
Affiliation 1:	Department of Biomedical Engineering, Faculty of Engineering, Islamic Azad University, Isfahan, Iran

Name 2:	Farnaz hesamifar - Corresponding
Affiliation 2:	Department of Biomedical Engineering, Central Tehran Branch, Islamic Azad University, 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 :

Accurate and reliable epilepsy detection, a common neurological disorder, remains a key challenge in clinical practice. Epileptic patients experience recurrent seizures caused by abnormal brain electrical activity, which disrupts neuronal communication and impairs consciousness. Timely and correct diagnosis can prevent serious complications and improve patient outcomes. Electroencephalography (EEG), a non-invasive technique for recording brain electrical activity, is widely used for functional diagnosis.In this study, two EEG signal processing frameworks were compared to classify three states: healthy, interictal, and ictal. Data from the well-known Bonn University dataset were preprocessed with filtering and normalization, then segmented into 2-second overlapping windows (50%) for training. The first framework extracted temporal and frequency-domain features and used a Random Forest classifier. The second framework applied a convolutional neural network (CNN) directly to raw EEG signals.Results showed that the Random Forest approach achieved 90.2% accuracy, while the CNN model outperformed it with 96.8% accuracy. These findings indicate that the proposed CNN, by learning directly from raw EEG, provides a more accurate and reliable method for automated epilepsy detection and holds strong potential for real-time clinical applications.

Methods:

In this study, the University of Bonn EEG database was used. First, a Butterworth band-pass filter with a passband of 0.5 to 45 Hz was applied. This frequency range covers the main EEG signal bands and was selected in accordance with previous studies. Next, each signal segment was independently normalized to have zero mean and unit variance. Due to the non-stationary nature of EEG signals, sliding windows with a length of 2 seconds and 50% overlap were employed. After preprocessing, a set of time-domain and frequency-domain features was extracted from each window. These extracted features were then used as inputs to a Random Forest classifier consisting of 100 decision trees with the Gini impurity criterion. To further improve detection performance, a single-layer Convolutional Neural Network (CNN) was designed and implemented. The network architecture included multiple layers with one-dimensional filters, pooling layers, regularization layers to prevent overfitting, and finally fully connected layer

Results:

The Random Forest model, after extracting time-domain and frequency-domain features from two-second windows, was able to discriminate among the three classes with an accuracy of 90.2%. The highest classification error occurred in distinguishing between the “healthy” and “interictal” classes, which may be attributed to the similarity of signal patterns in these two conditions. The Convolutional Neural Network (CNN) model, which was trained end-to-end on raw signal windows, demonstrated superior performance. After completing the training process, this model achieved an overall accuracy of 96.8% on the test set.

Conclusion:

A quantitative comparison of the results obtained from the two methods clearly confirms the superiority of the deep learning approach. This performance improvement can be attributed to the inherent capability of deep networks to learn complex and hierarchical representations directly from raw data. These results conclusively demonstrate that the proposed Convolutional Neural Network architecture provides a more accurate and reliable solution for the automatic detection of epilepsy from electroencephalogram (EEG) signals.

Keywords:

Epilepsy, Electroencephalogram (EEG), Random Forest, Convolutional Neural Network, Deep Learning