CNS*2026 Halifax

Introduction
Recent work highlights the importance of identifying and assessing sequential neural activity. Here we introduce a novel approach to characterize sequential patterns in EEG data, focusing on steady-state visually evoked potentials (SSVEPs) that appear as oscillations in the occipital lobe when attending to a flickering light. This phenomenon is widely used in brain-computer interfaces [1], and is often characterized solely in terms of oscillations and their frequency components, a framework that may overlook critical aspects of neural dynamics [2]. Using spectrograms and wavelet transforms, we examine the temporal evolution of frequency and power in an SSVEP BCI speller dataset and assess its relationship with BCI performance.

Methods


The EEG recordings used in our analysis come from the BETA dataset [3], which includes recordings from 70 subjects in a BCI speller experiment with 40 different stimulus frequencies and 4 trials per frequency. We used spectrograms and wavelet transforms to examine the evolution of the signal’s frequency content and power over time. In this analysis, we aimed to assess how long a response attributable to the stimulus frequency is maintained throughout the trial, as well as how long it takes for the response to appear (latency). These results were obtained for all subjects, averaged across trials, and compared with their respective performance using a traditional canonical correlation analysis (CCA) detection method.

Results
After characterizing the EEG signals in terms of three metrics (active time percentage (ATP) in the wavelet transform, active time percentage (ATP) in the spectrogram, and response latency), we computed the average results for each subject across four trials.  We then examined these results and compared them with each subject’s CCA performance. We found a highly significant positive correlation between the CCA performance and the two metrics evaluating the percentage of active time: CWT ATP (r = 0.896), and Spectrogram ATP (r = 0.943). On the other hand, we found a highly significant negative correlation with the response latency metric (r = –0.907).

Discussion
Our characterization of the evolving response reveals that SSVEPs are not universally steady as their name suggests. High-performing subjects react to the stimulus earlier and tend to maintain a response throughout the trial.  In contrast, subjects with poor performance exhibit unstable responses with sequential activations across different frequency bands contributing to detection errors. We also found that response latency is linked to performance: a subject who responds correctly but reacts slowly will perform poorly in the BCI task. Our analysis suggests that poor performance may not always result from an inherent inability to respond to stimuli, but rather from visual attention issues in the context of simultaneous visual stimuli.

References
[1] Liu,  S., Zhang, D.,  Liu,  Z.,  Liu,  M.,  Ming,  Z.,  Liu,  T.,  Suo, D., Funahashi, S., and Yan, T. (2022).   Review of brain–computer in- terface based on steady-state visual evoked  potential.   Brain  Science Advances,   8(4): 258–275, https://doi.org/10.26599/BSA.2022.9050022.
[2] Labecki, M., Nowicka, M. M., Wrobel,  A., and Suffczynski, P. (2024). Frequency-dependent  dynamics of  steady-state visual  evoked pottials under sustained flicker stimulation.  Scientific Reports, 14(1):9281, https://doi.org/10.1038/s41598-024-59770-5.
[3] Liu, B., Huang, X., Wang, Y., Chen, X., and Gao, X. (2020). BETA: A Large Benchmark Database Toward SSVEP-BCI Application.  Frontiers in Neuroscience, 14, https://doi.org/10.3389/fnins.2020.00627.

Acknowledgement
Research funded by grants PID2024-155923NB-I00 and CPP2023-010818 (MCIN/AEI and ERDF- "A way of making Europe").