Multi-Class Classification of Upper Limb Movements With Filter Bank Task-Related Component Analysis.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Non-invasive brain-computer interfaces (BCIs) rely on classifying limb movements for control.
• Existing BCIs primarily focus on left/right limb classification, neglecting diverse upper limb movements.
• Advanced machine learning methods are needed for effective multi-class classification of limb movements.

Purpose of the Study:

• To develop and evaluate a novel method for multi-class classification of upper limb movements.
• To enhance the number of active-evoked control commands in non-invasive BCIs.
• To improve the accuracy of limb movement classification using electroencephalography (EEG) signals.

Main Methods:

• Proposed the multi-class filter bank task-related component analysis (mFBTRCA) method.
• Employed spatial filtering (task-related component analysis) to reduce noise in EEG signals.
• Utilized canonical correlation for feature extraction and minimum-redundancy maximum-relevance for feature selection, followed by support vector machine classification.

Main Results:

• The mFBTRCA method achieved classification accuracies of 0.4193 ± 0.0780 (7 classes) and 0.4032 ± 0.0714 (5 classes).
• These results represent a significant improvement over existing methods, which achieved accuracies of 0.3590 ± 0.0645 and 0.3159 ± 0.0736, respectively.
• The method demonstrated superior performance in multi-class upper limb movement classification.

Conclusions:

• The proposed mFBTRCA method effectively classifies multiple types of upper limb movements from EEG signals.
• This advancement is expected to enable more sophisticated control commands in non-invasive BCIs.
• The study highlights the potential of advanced machine learning for improving BCI functionality.