Visual Brain-Computer Interface Based on High-Frequency Steady-State Visual Evoked Fields

Background

Brain-Computer Interface (BCI) technology allows users to control machines by decoding specific brain activity signals. While invasive BCIs excel in capturing high-quality brain signals, their application is mainly limited to clinical settings. Non-invasive methods, such as Electroencephalography (EEG), provide a more feasible pathway for broader BCI applications. However, due to the influence of cerebrospinal fluid and the skull, EEG signals become very weak during transmission, and the varying conductivity and anisotropy of the skull make locating EEG signal sources more difficult.

Magnetoencephalography (MEG) is a non-invasive method for imaging brain activity, superior to EEG in capturing fine spatial information. This advantage primarily arises because magnetic flux does not attenuate like electrical currents. However, traditional MEG equipment using Superconducting Quantum Interference Devices (SQUIDs) placed 3-6 cm from the scalp results in low Signal-to-Noise Ratio (SNR), and the equipment requires continuous cryogenic cooling, increasing cost and operational constraints. Recently, Optically Pumped Magnetometers (OPM) have been introduced into MEG measurement as an emerging technology. OPMs are smaller and do not need cooling, theoretically providing sensitivity comparable to SQUIDs.

Steady-State Visual Evoked Potentials (SSVEPs) stand out among EEG features and are often used as control signals in various BCI systems. Most SSVEP BCI systems use low-frequency (<12 Hz) or medium-frequency (12–30 Hz) stimuli, which produce stronger signals. However, low and medium-frequency SSVEPs can impair users’ actual experience and even increase the risk of visual fatigue and epilepsy. In contrast, high-frequency SSVEP stimulation provides a more comfortable interactive experience, becoming imperceptible when the flicker frequency exceeds the critical fusion threshold (50-60 Hz), thus greatly reducing fatigue and enhancing user experience. However, due to the poor performance of high-frequency SSVEP-BCI systems, they have seen less practical application.

To address this challenge, this study constructs an OPM-MEG-based BCI system using high-frequency Steady-State Visual Evoked Fields (SSVEFs) to explore its feasibility.

Source of the Paper

This research paper is authored by Dengpei Ji, Xiaolin Xiao, Jieyu Wu, Xiang He, Guiying Zhang, Ruihan Guo, Miao Liu, Minpeng Xu, Qiang Lin, Tzyy-Ping Jung, Dong Ming, researchers from the School of Medicine and Engineering Interdisciplinary and Translational Medicine at Tianjin University, the School of Science at Zhejiang University of Technology, the University of California, San Diego, and the Haihe Laboratory of Brain-Computer Interaction and Human-Machine Integration in Tianjin, China. The paper was published on May 30, 2024, in the Journal of Neural Engineering. The abstract highlights the key point of their research: constructing an MEG-BCI system based on high-frequency SSVEFs to achieve imperceptible flicker, user-friendliness, and high accuracy.

Research Methods

Experimental Equipment and Environment

The research team used six second-generation near-zero field optically pumped magnetometers manufactured by QuSpin (Louisville, Colorado, USA), each a self-contained sensor unit capable of measuring magnetic fields in two orthogonal directions. The sensors were placed in a zero-magnetic field shielded room to meet the no-magnetic-field environment requirements, reducing the ambient static magnetic field to about 2 nanoteslas, with a gradient not exceeding 15 nanoteslas per meter.

Participants wore standard ¹⁰⁄₂₀ EEG caps, with helmets fixing the sensor positions to ensure accurate signal acquisition.

Subjects

Five subjects (aged between 18 and 30, including one female, all right-handed) participated in the experiment. All participants confirmed no claustrophobia and fully understood the experimental process before the experiment.

Stimulus Presentation

The research team designed a high-frequency SSVEF system with a 9-command closed array, ranging 58-62 Hz in frequency with 0.5 Hz intervals, driven by white LEDs controlled by Cyclone IV Field-Programmable Gate Array (FPGA). Optical fibers were used to transmit the stimulation signals into the shielded room to meet no-magnetic-field environment requirements.

Experimental Procedure

Each subject completed 15 sets of offline experiments, with each set containing 3 blocks, and each block containing 9 stimuli presented in pseudorandom order. Each stimulus lasted 4 seconds, totaling 135 seconds. To mitigate visual fatigue in the dark environment, the stimulus unit remained continuously illuminated during breaks.

Data Recording and Analysis

OPM-MEG systems recorded data, with FPGA sending trigger signals for stimuli, using built-in signal processing software for preprocessing. Infinite Impulse Response band-pass filters were set to a range of 55-70 Hz, using Fast Fourier Transform (FFT) for frequency domain characteristics analysis, and employing Ensemble Task-Related Component Analysis (ETRCA) algorithm for target recognition and system performance evaluation.

Key Research Results

Environmental Noise Characteristics Analysis

By analyzing the 4-70 Hz spectrum energy of subjects in a quiet state, it was confirmed that low-frequency noise does not affect high-frequency SSVEF signal detection, verifying the normal detection capability of MEG signals.

Time-Domain Characteristics of High-Frequency SSVEF

Analyzing the waveform from 200 ms before to 400 ms after stimulation onset confirmed that signals were stable with significant amplitude and phase after a visual latency of approximately 200 ms.

Frequency-Domain Analysis of High-Frequency SSVEF

FFT analysis showed significant energy peaks at target frequencies for each event, reflecting the successful recognition of MEG signals. The z-axis energy was higher than the y-axis in all events, showing significant differences at different target frequencies (p<0.01).

Performance of High-Frequency SSVEF-BCI System

In offline experiments, the nine-command BCI system achieved an average classification accuracy of 92.98%, with a theoretical maximum Information Transfer Rate (ITR) of 58.36 bits/min, and the longest individual ITR reached 63.75 bits/min (Subject 3), indicating a high ITR level with a short data length (0.7 seconds).

Research Conclusion

This study explored the feasibility of an OPM-MEG BCI system based on high-frequency SSVEFs for the first time, achieving a significant average offline accuracy (92.98%) and an impressive maximum ITR (63.75 bits/min). The results demonstrate the potential and feasibility of MEG in detecting weak brain signals, providing theoretical and practical value for advancing MEG’s development and application in BCI systems.

Highlights of the Experiment

1. High Innovation: This is the first exploration of an OPM-MEG BCI system based on high-frequency SSVEFs, addressing visual fatigue and epilepsy risk associated with traditional low-frequency SSVEFs.
2. High Accuracy and ITR: The system exhibited excellent performance with an average classification accuracy of 92.98% and a theoretical maximum ITR of 58.36 bits/min.
3. Multi-Dimensional Signal Analysis: Joint analysis of z-axis and y-axis signals significantly improved system classification accuracy, showcasing the potential of multi-dimensional analysis methods.

Future Research Directions

Future research will focus on expanding the number of commands and improving algorithms suitable for weak MEG signals to further enhance ITR. Other potential applications include more online experiments with a larger number of subjects to demonstrate system universality and comparing different characteristics in joint MEG and EEG experiments, exploring broader applications of MEG in BCI systems.