Related Experiment Video
Updated: Jan 6, 2026
08:51
Statistical Modelling of Cortical Connectivity Using Non-invasive Electroencephalograms
Published on: November 1, 2019
6.0K
Percept-related EEG classification using machine learning approach and features of functional brain connectivity
Alexander E Hramov1, Vladimir Maksimenko1, Alexey Koronovskii2
1Neuroscience and Cognitive Technology Laboratory, Center for Technologies in Robotics and Mechatronics Components, Innopolis University, 420500 Innopolis, The Republic of Tatarstan, Russia.
Chaos (Woodbury, N.Y.)
|October 3, 2019
Summary
Machine learning effectively classifies electroencephalographic (EEG) trials by identifying neurophysiological features. Focusing on brain network properties significantly improves classification accuracy, especially for complex cognitive tasks.
Area of Science:
- Neuroscience
- Machine Learning
- Signal Processing
Background:
- Machine learning (ML) is crucial for electroencephalographic (EEG) trials classification.
- Feature extraction and selection are key to reducing data dimensionality and improving ML efficiency.
- Traditional dimensionality reduction methods often overlook the neurophysiological origins of EEG data.
Purpose of the Study:
- To investigate if EEG features, rooted in neurophysiological processes, possess distinct spatiotemporal characteristics.
- To leverage prior knowledge of neurophysiology for selecting principal EEG features.
- To enhance EEG trial classification accuracy by considering network properties and optimizing ML models.
Main Methods:
- Classifying EEG trials related to ambiguous visual stimuli perception.
- Analyzing functional neural interactions to identify brain areas with distinct network architectures.
- Optimizing a feedforward multilayer perceptron and developing a training set selection strategy.
- Evaluating classification accuracy with full and reduced channel sets.
Main Results:
- EEG features associated with different interpretations of ambiguous stimuli are linked to neuronal network properties.
- Specific brain areas showing differences in neural network architecture were identified.
- Classification accuracy reached 85% using all channels and improved to approximately 95% with optimized feature selection and reduced channels (up to 90% reduction).
Conclusions:
- EEG features exhibit distinct spatiotemporal characteristics tied to underlying neurophysiological processes.
- Incorporating knowledge of brain network topology and functional interactions enhances EEG classification.
- The proposed method offers a pathway for accurate EEG classification in complex cognitive tasks.
