A hierarchical architecture for recognising intentionality in mental tasks on a brain-computer interface.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCI) rely on electroencephalography (EEG) signals for user intent detection.
• Accurately distinguishing intentional control (IC) from non-intentional control (NC) states is crucial for reliable BCI operation.
• Minimizing the false positive rate (FPR) in NC classification is essential to prevent unintended actions.

Purpose of the Study:

• To propose a hierarchical machine learning system for recognizing intentional and non-intentional mental tasks in BCI.
• To develop a method for automatic detection of user intent within EEG signals.
• To classify intended commands from recognized intentional states.

Main Methods:

• A hierarchical system employing machine learning techniques on EEG signals.
• First-level clustering to differentiate between IC and NC states.
• Supervised learning techniques applied to IC patterns for command classification.
• Targeting a maximum FPR of 10% for NC states.

Main Results:

• The proposed system achieved an average test accuracy of 66.6% on the BCI competition IIIa dataset.
• A low FPR of 8.2% was obtained under the specified conditions.
• Demonstrated the effectiveness of the hierarchical approach in discriminating between intentional and non-intentional states.

Conclusions:

• The hierarchical machine learning approach provides an effective method for intentional and non-intentional state discrimination in BCI.
• The system successfully classifies intended commands while maintaining a low FPR, crucial for safe BCI applications.
• This study contributes a novel strategy for enhancing the reliability and usability of BCI systems.