Adaptive LDA Classifier Enhances Real-Time Control of an EEG Brain-Computer Interface for Decoding Imagined...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-Computer Interfaces (BCIs) offer communication pathways for individuals with speech impairments by decoding neural signals.
• Traditional BCIs use static classifiers, which struggle with the non-stationary nature of electroencephalography (EEG) and user learning.
• Real-time adaptation is crucial for improving BCI accuracy and usability.

Purpose of the Study:

• To develop and evaluate an adaptive classifier for real-time decoding of imagined speech using EEG signals.
• To compare the performance of an adaptive Linear Discriminant Analysis (LDA) classifier against a static LDA classifier.

Main Methods:

• An adaptive LDA classifier was developed, optimizing parameters like the update coefficient (UC) using prior EEG data.
• The adaptive and static LDA classifiers were tested in a real-time BCI control task involving imagined syllable decoding.
• Twenty healthy participants engaged in two BCI control sessions, using both classifier types in a randomized order.

Main Results:

• The adaptive LDA classifier demonstrated superior performance in the real-time BCI control task compared to the static LDA classifier.
• Optimal parameters for the adaptive classifier showed consistency across different datasets from the same task.
• The adaptive approach effectively addressed the non-stationarity of EEG signals and user adaptation.

Conclusions:

• Adaptive LDA classifiers are effective and reliable for real-time imagined speech decoding.
• This advancement can reduce training time and facilitate the development of more sophisticated multi-class BCIs.
• The findings are particularly relevant for non-invasive BCIs that often face challenges with low decoding accuracy.