Related Experiment Video
Updated: Aug 4, 2025
07:42
Combined Transcranial Magnetic Stimulation and Electroencephalography of the Dorsolateral Prefrontal Cortex
Published on: August 17, 2018
11.9K
Effect of time windows in LSTM networks for EEG-based BCIs
K Martín-Chinea1, J Ortega1, J F Gómez-González1
1Department of Industrial Engineering, University of La Laguna, 38071 San Cristóbal de La Laguna, Tenerife Spain.
Cognitive Neurodynamics
|April 3, 2023
Summary
This study demonstrates that Long Short-Term Memory (LSTM) networks significantly improve the accuracy of brain-computer interfaces (BCIs) for individuals with motor impairments. Using real-time electroencephalogram (EEG) data, LSTMs offer a safer and more reliable control method for assistive technologies like smart wheelchairs.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Biomedical Engineering
Background:
- Brain-computer interfaces (BCIs) offer potential for individuals with motor impairments.
- Current EEG-based BCIs lack the accuracy and safety required for real-world applications, especially in dynamic environments.
- Challenges include low signal-to-noise ratios and signal contamination in portable EEG systems.
Purpose of the Study:
- To evaluate the effectiveness of Long Short-Term Memory (LSTM) networks for real-time EEG signal interpretation in BCIs.
- To determine the optimal time window for LSTM classification to maximize accuracy.
- To assess the feasibility of implementing an LSTM-based BCI in a smart wheelchair for individuals with reduced mobility.
Main Methods:
- Utilized a low-cost wireless electroencephalogram (EEG) device for real-time data acquisition.
- Applied Long Short-Term Memory (LSTM) networks, a type of recurrent neural network, to classify user intentions from EEG signals.
- Investigated various time windows to identify the optimal duration for classification accuracy.
- Conducted tests in real-life contexts to evaluate performance and necessary trade-offs.
Main Results:
- LSTM networks achieved significantly higher classification accuracy (77.61%–92.14%) compared to traditional classifiers (59.71%).
- An optimal time window of approximately 7 seconds was identified for the specific tasks studied.
- Real-life tests highlighted the necessity of balancing accuracy and response time for reliable detection.
Conclusions:
- LSTM networks demonstrate superior performance in interpreting EEG signals for BCI applications.
- The developed LSTM-based BCI shows promise for enhancing the safety and functionality of smart wheelchairs for users with motor impairments.
- Further research should focus on optimizing the accuracy-response time trade-off for robust real-world deployment.
