Burst c-VEP Based BCI: Optimizing Stimulus Design for Enhanced Classification with Minimal Calibration Data and Improved User Experience

Abstract

The Steady State Visual Evoked Potential (SSVEP) is a widely used technique in Brain-Computer Interface (BCI) research due to its high information transfer rate. However, this method has some limitations, including lengthy calibration time and visual fatigue. Recent studies have explored the use of code-modulated Visual Evoked Potentials (c-VEP) with aperiodic flickering visual stimuli as an alternative approach to address these issues. One advantage of c-VEP is that the training of the model is independent of the number and complexity of targets, which helps reduce calibration time. Nevertheless, the existing designs of c-VEP can be further improved to achieve a higher signal-to-noise ratio, and shorten the selection time and the calibration process. In this study, we propose a novel type of code-VEP called "Burst c-VEP" that consists of brief presentations of aperiodic visual flashes at approximately 3Hz. The rationale behind this design is to leverage the sensitivity of the primary visual cortex to transient changes in low-level stimuli features to reliably elicit distinctive series of visual evoked potentials. In comparison to other types of faster-paced codes, burst c-VEP exhibits favorable properties to achieve high decoding performance using convolutional neural networks (CNN). We also explore the attenuation of visual stimuli contrast and intensity to further reduce the perceptual saliency of c-VEP. The proposed solutions were tested through an offline 4-classes c-VEP protocol involving 12 participants. Following a factorial design, participants were asked to focus on c-VEP targets whose pattern (burst and maximum-length sequence) and amplitude (100% or 40% amplitude depth modulation) were manipulated across experimental conditions. Firstly, the full amplitude burst c-VEP codes exhibited higher accuracy ranging from 90.5% (with 17.6𝑠 of calibration data) to 95.6% (with 52.8𝑠 of calibration data) than its m-sequence counterpart (71.4% to 85.0%). The mean selection time for the two types of codes (1.5𝑠) compared favorably compared to existing studies. Secondly, our findings revealed that lowering the intensity of the stimuli barely decreased the accuracy of the burst to 94.2% accuracy while yielding a higher subjective visual user experience. Taken together, these results demonstrate the high potential of the proposed burst codes to advance BCI beyond the confines of the lab. The collected datasets, along with the proposed CNN architecture implementation, are shared through open-access repositories.