Enhancing detection of SSVEPs using discriminant compacted network.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Steady-state visual evoked potential-based brain-computer interfaces (SSVEP-BCIs) offer high signal-to-noise ratios and information transfer rates (ITRs).
• Accurate detection of SSVEPs is crucial for improving SSVEP-BCI system performance.
• Existing methods often struggle with optimal feature extraction and classification accuracy.

Purpose of the Study:

• To introduce a novel decoding method, Discriminant Compacted Network (Dis-ComNet), for enhanced SSVEP detection.
• To leverage both spatial filtering and deep learning (DL) for improved feature extraction.
• To evaluate the performance of Dis-ComNet across diverse SSVEP datasets.

Main Methods:

• Dis-ComNet utilizes global template alignment and discriminant spatial patterns to enhance SSVEP features.
• A compacted temporal-spatio module (CTSM) is designed for finer feature extraction.
• The method was validated on high-frequency, Benchmark, and wearable SSVEP datasets.

Main Results:

• Dis-ComNet significantly outperformed state-of-the-art spatial filtering and deep learning methods on all tested datasets.
• Classification accuracy improvements ranged from 2.5% to 37.5% compared to various existing methods.
• Achieved information transfer rates (ITRs) reached up to 126.0 bits/min, 236.4 bits/min, and 103.6 bits/min for the respective datasets.

Conclusions:

• Dis-ComNet demonstrates superior performance in SSVEP detection compared to current methods.
• The proposed method effectively enhances SSVEP features and extracts finer details.
• This development facilitates the creation of high-accuracy SSVEP-BCI systems.