Improving Classification Performance of a brain-Computer Interface System based on Rapid Serial Visual Presentation by Shifting stimuli

Dong-Ok Won, Han-Jeong Hwang, Klaus-Robert Müller, Seong-Whan LeeGBCIC 2017 · OpenReview

Abstract

Brain-Computer Interface (BCI) spellers based on Rapid Serial Visual Presentation (RSVP) provide a critical communication tool for "locked-in" patients who have lost eye movement control. However, traditional static RSVP (S-RSVP) paradigms are often limited by low accuracy and high mental fatigue. This research proposes a motion-based RSVP (M-RSVP) paradigm that introduces a directional shift to visual stimuli. By moving characters from a central point toward the periphery within the foveal vision, the system leverages the human visual system's sensitivity to motion. Results show that M-RSVP achieves a mean classification accuracy of 92.36%, a substantial improvement over the 73.61% achieved by S-RSVP.

Introduction

Most BCI spellers rely on the user's ability to perform eye movements to select letters on a grid. For patients in the late stages of Amyotrophic Lateral Sclerosis (ALS), who may lose all oculomotor control, gaze-independent BCIs are essential. RSVP is a common gaze-independent method where letters flash sequentially in a single central location. Despite its utility, static RSVP is mentally taxing and often results in poor signal-to-noise ratios in EEG data. This study explores whether adding a motion component to these flashes can enhance the brain's response, making target letters easier to distinguish and improving overall system performance.

Methodology

The researchers compared two paradigms: Standard Static RSVP (S-RSVP) and Motion RSVP (M-RSVP). In S-RSVP, 36 characters (A-Z, 1-9, and a hyphen) were flashed one-by-one in the center of the screen. In M-RSVP, characters appeared centrally but immediately moved toward one of six radial directions (2, 4, 6, 8, 10, and 12 o'clock) within a visual angle of 2.89°. This ensures the stimuli remain within the user's foveal vision, maintaining true gaze independence.

Comparison of (a) S-RSVP and (b) M-RSVP paradigms. In M-RSVP, characters shift from the center toward one of six directions.

The study involved eight healthy participants. EEG signals were recorded using a 63-channel system at 1000 Hz. The Stimulus Onset Asynchrony (SOA) was set to 133ms. Classification was performed using Regularized Linear Discriminant Analysis (RLDA) with shrinkage, targeting specific Event-Related Potentials (ERPs) such as the N200 and P300 waves.

Physiological Evidence and Results

The core finding of this study is that motion acts as a neural amplifier. The M-RSVP paradigm produced ERPs with significantly higher amplitudes and shorter latencies compared to S-RSVP. Specifically, the N200 wave—associated with visual processing—appeared at 225ms for M-RSVP, nearly 100ms faster than the 315ms observed in S-RSVP.

ERP waveforms for M-RSVP at channels PO7 and Cz, highlighting the N200 and P300 components used for classification.

These sharper neural signatures translated directly into better machine learning performance. After 10 sequences, M-RSVP reached a mean accuracy of 92.36%, while S-RSVP reached 73.61%. Even with fewer sequences (which allows for faster typing), M-RSVP maintained a significant lead, achieving 81.74% accuracy after only 6 sequences.

Discussion and Future Work

While the results are promising, the study was conducted with a small sample of healthy participants. Future research must validate these findings with the target demographic—patients with severe motor and oculomotor impairments. Additionally, moving from offline analysis to real-time "online" testing is necessary to confirm the system's practical utility. The authors also plan to investigate more advanced classification techniques, such as non-linear regression and spectral feature analysis, to further optimize performance.

Summary updated

BibTeX
@inproceedings{won2017improving,
  author = {Dong-Ok Won and Han-Jeong Hwang and Klaus-Robert Müller and Seong-Whan Lee},
  title = {{Improving Classification Performance of a brain-Computer Interface System based on Rapid Serial Visual Presentation by Shifting stimuli}},
  booktitle = {GBCIC},
  year = {2017},
  url = {https://openreview.net/forum?id=LhvLzZRrhCE},
}
Figure