A Voting-Enhanced Dynamic-Window-Length Classifier for SSVEP-Based BCIs.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Brain-computer interfaces (BCIs) offer communication pathways for individuals with severe motor impairments.
• Steady-state visual evoked potential (SSVEP)-based BCIs are a common BCI paradigm.
• Existing SSVEP BCIs often require complex user-specific calibration and feature selection.

Purpose of the Study:

• To develop a novel, unsupervised, dynamic window-length classifier for SSVEP-based BCIs.
• To enhance ease-of-use by eliminating the need for feature extraction method or channel set selection.
• To improve the performance of SSVEP BCIs, particularly information transfer rates (ITR).

Main Methods:

• A dynamic window-length classifier was developed, employing multiple feature extraction methods and channel selections.
• Majority voting was used to infer SSVEP targets, with dynamic window extension if no majority was reached.
• The classifier was evaluated on a public dataset from 35 healthy participants and compared against MEC, MSI, and FBCCA.

Main Results:

• The proposed classifier demonstrated superior performance, achieving an average ITR of 123.5 bits-per-minute (bpm).
• This represents a significant increase in ITR compared to existing methods: 47.5 bpm over MEC, 51.2 bpm over MSI, and 19.5 bpm over FBCCA.
• The classifier is unsupervised, adapting channel selection and using dynamic window lengths without user training.

Conclusions:

• The dynamic window-length classifier offers a user-friendly and effective approach to SSVEP-based BCIs.
• Its unsupervised nature and adaptive features make it particularly suitable for caregivers and users with limited technical expertise.
• The method significantly enhances BCI performance, paving the way for more accessible and efficient assistive technologies.