Toward a Subject-Independent EEG-Based Neural Indicator of Task Proficiency During Training

Abstract

This study explores the feasibility of developing an EEG-based neural indicator of task proficiency based on subject-independent mental state classification. Such a neural indicator could be used in the development of a passive brain-computer interface to potentially enhance training effectiveness and efficiency. A spatial knowledge acquisition training protocol was used in this study. Fifteen participants acquired spatial knowledge in a novel virtual environment via 60 navigation trials (divided into ten blocks). Task performance (time required to complete trials), perceived task certainty, and EEG signal data were collected. For each participant, 1 s epochs of EEG data were classified as either from the "low proficiency, 0" or "high proficiency, 1" state using a support vector machine classifier trained on data from the remaining 14 participants. The average epoch classification per trial was used to calculate a neural indicator (NI) ranging from 0 ("low proficiency") to 1 ("high proficiency"). Trends in the NI throughout the session-from the first to the last trial-were analyzed using a repeated measure mixed model linear regression. There were nine participants for whom the neural indicator was quite effective in tracking the progression from low to high proficiency. These participants demonstrated a significant (p < 0.001) increase in the neural indicator throughout the training from NI = 0.15 in block 1 to NI = 0.81 (on average) in block 10, with the average NI reaching a plateau after block 7. For the remaining participants, the NI did not effectively track the progression of task proficiency. The results support the potential of a subject-independent EEG-based neural indicator of task proficiency and encourage further research toward this objective.

Keywords: EEG; classification; cognitive state prediction; passive brain-computer interface; spatial knowledge acquisition; subject-independent.

Figure 1

Electrode placement map with electrode names and cortical region labels (frontal, temporal, parietal, and occipital). Adapted from (Kenny et al., 2019).

Figure 2

Time performance (%) per trial for each participant, calculated as the percent difference from an “ideal” time performance.

Figure 3

Perceived certainty rating (1–10) per trial for each participant.

Figure 4

Mean neural indicator (NI) per trial for each participant. Dotted lines represent the 95% confidence interval. NI = 0 (“low proficiency”), NI = 1 (“high proficiency”). *Participants for which the NI was effective.

Figure 5

Mean neural indicator (NI) per block distinguishing the two subgroups for which the NI was effective and ineffective, as well as for all participants combined. NI = 0 (“low proficiency”), NI = 1 (“high proficiency”).

Figure 6

Histogram of features selected during classification. Each plot displays the absolute frequency of feature selection for features in the alpha-high frequency band organized by brain region (frontal, parietal, temporal, and occipital).