Related Experiment Video
Updated: Sep 27, 2025
08:51
Statistical Modelling of Cortical Connectivity Using Non-invasive Electroencephalograms
Published on: November 1, 2019
5.8K
Graphical representation learning-based approach for automatic classification of electroencephalogram signals in
Surbhi Soni1, Ayan Seal1, Anis Yazidi2
1PDPM Indian Institute of Information Technology, Design and Manufacturing, Jabalpur, 482005, India.
Computers in Biology and Medicine
|April 7, 2022
Summary
This study introduces an automated method using Electroencephalogram (EEG) data and graph-based embeddings for accurate depression detection. The novel approach achieves high accuracy, simplifying feature extraction for mental health diagnostics.
Area of Science:
- Neuroscience
- Computational Psychiatry
- Machine Learning
Background:
- Depression, a major depressive disorder, significantly impacts global health and well-being.
- Early detection of depression is crucial to mitigate its adverse physical and emotional consequences.
- Electroencephalogram (EEG) offers a non-invasive method for analyzing brain activity to identify depression.
Purpose of the Study:
- To develop an automated feature extraction method for depression detection using EEG data.
- To leverage graph construction and Node2vec for generating node embeddings from subject relationships.
- To evaluate the efficacy of different data fusion techniques (graph-level, feature-level, decision-level) for enhanced diagnostic accuracy.
Main Methods:
- Constructing a graph where nodes represent subjects and edge weights signify relationships (Euclidean distance).
- Employing the Node2vec framework to generate node embeddings, preserving similarity in the graph.
- Implementing three fusion methods (graph-level, feature-level, decision-level) to combine multi-channel EEG features.
Main Results:
- The proposed method demonstrated effective depression detection across three public EEG datasets.
- Decision-level fusion achieved a peak accuracy of 0.933, outperforming existing state-of-the-art methods.
- The automated feature extraction significantly reduced the need for manual feature engineering.
Conclusions:
- The automated graph-based feature extraction and fusion method provides a robust approach for EEG-based depression detection.
- This technique offers a promising advancement in computational psychiatry for mental health diagnostics.
- The Node2vec embeddings and fusion strategies enhance the accuracy and efficiency of identifying depression from brain signals.
