An exploration of BCI performance variations in people with amyotrophic lateral sclerosis using longitudinal EEG data

Abstract

Objective: Brain-computer interface (BCI) technology enables people to use direct measures of brain activity for communication and control. The National Center for Adaptive Neurotechnologies and Helen Hayes Hospital are studying long-term independent home use of P300-based BCIs by people with amyotrophic lateral sclerosis (ALS). This BCI use takes place without technical oversight, and users can encounter substantial variation in their day-to-day BCI performance. The purpose of this study is to identify and evaluate features in the electroencephalogram (EEG) that correlate with successful BCI performance during home use with the goal of improving BCI for people with neuromuscular disorders.

Approach: Nine people with ALS used a P300-based BCI at home over several months for communication and computer control. Sessions from a routine calibration task were categorized as successful ([Formula: see text]70%) or unsuccessful (<70%) BCI performance. The correlation of temporal and spectral EEG features with BCI performance was then evaluated.

Main results: BCI performance was positively correlated with an increase in alpha-band (8-14 Hz) activity at locations PO8, P3, Pz, and P4; and beta-band (15-30 Hz) activity at occipital locations. In addition, performance was significantly positively correlated with a positive deflection in EEG amplitude around 220 ms at frontal mid-line locations (i.e. Fz and Cz). BCI performance was negatively correlated with delta-band (1-3 Hz) activity recorded from occipital locations.

Significance: These results highlight the variability found in the EEG and describe EEG features that correlate with successful BCI performance during day-to-day use of a P300-based BCI by people with ALS. These results should inform studies focused on improved BCI reliability for people with neuromuscular disorders.

Figure 1.

Performance variations for six representative users. The y -axis presents classification accuracy (as % correct), and the x-axis presents the number of days after the initial session. Each connected data point indicates the mean accuracy for one day. The vertical bars reflect the minimum and maximum performance for that day. Data were not available for the days without vertical bars. The red dashed line indicates the linear performance trend over time.

Figure 2.

Average topographies of significant thresholded correlations between online accuracy and spectral power. The correlations corresponding to p ⩾ 0.05 (with Bonferroni correction) were set to zero as not significant, and the significant positive and negative correlations were set to 1 and −1, respectively.

Figure 3.

Average topographies of significant thresholded correlations between online accuracy and N100, N200, and 500–700 ms minimum peak amplitude and P200 and P300 maximum peak amplitude. The correlations corresponding to p ⩾ 0.05 (with Bonferroni correction) were set to zero as not significant, and the significant positive and negative correlations were set to 1 and −1, respectively.

Figure 4.

The normalized power spectra for successful (blue) versus unsuccessful (red) runs averaged over all subjects. The solid gray bars indicate the statistically significant features (p <0.05, with Bonferroni correction).

Figure 5.

The normalized target ERPs for successful (blue) versus unsuccessful (red) runs averaged over all subjects. The solid gray bars indicate the statistically significant features (p <0.05, with Bonferroni correction).