EEG-Based Multiword Imagined Speech Classification for Persian Words.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) enable communication and control through neural signals.
• Electroencephalography (EEG) offers a non-invasive method for capturing brain activity.
• Developing accurate classifiers for imagined speech is crucial for advancing BCIs.

Purpose of the Study:

• To create a simple, extensible, multiclass classifier for imagined Persian words using EEG signals.
• To evaluate the classifier's performance across different temporal modes (long-time, short-time, mixed-time).
• To demonstrate the feasibility of using imagined words for controlling devices or computer interfaces.

Main Methods:

• Utilized EEG signals from 19 channels, focusing on frequency spectrums (1-32 Hz).
• Employed a majority rule of binary Support Vector Machine (SVM) classifiers for multiclass classification.
• Implemented Monte-Carlo cross-validation to estimate accuracy and confusion matrices.
• Defined three classification modes based on temporal data separation: long-time, short-time, and mixed-time.

Main Results:

• Achieved accuracies ranging from 32% (7-class, long-time) to 97% (mixed-time).
• Short-time and mixed-time modes significantly outperformed the long-time mode, with accuracies up to 97%.
• Even the long-time mode's results were significantly better than random chance and comparable to existing studies.

Conclusions:

• The proposed EEG-based classifier for imagined words is effective and extensible.
• Performance is highly dependent on the temporal proximity of training and testing data.
• The findings support the potential of this approach for practical BCI applications like device control.