Compact EEG Tracks Mental Workload Across Varying Task Difficulty and Stimulus Pacing

Compact electroencephalography recordings capture condition-associated neural signatures across working memory tasks, but their decoding performance remains modest and highly context-dependent, according to an original research study conducted by Giuseppe Spaltro, Luca Serioli, Sebastian Sieghartsleitner, Johannes Gruenwald, Christoph Guger, and Arcangelo Merla. Investigating the feasibility of mental workload monitoring for adaptive systems, the researchers evaluated whether low-density brain monitoring could reliably track variations across different task structures, operational durations, and individual users while relying on a feasible eight-channel sensor layout.

The experimental protocol evaluated 15 healthy adult participants performing two complementary paradigms based on the N-back working memory framework. The first manipulation, designated N-LEVELS, varied cognitive memory load across 1-back, 2-back, and 3-back difficulty tiers. The second manipulation, designated N-SPEED, held cognitive load constant at the 2-back level while systematically altering the stimulus pacing across slow, medium, and fast presentation speeds. Electroencephalography signals were gathered throughout these sessions using a compact eight-channel configuration to assess the viability of low-density sensor layouts in cognitive neuroergonomics.

Behavioral measurements and subjective self-reports demonstrated distinct profiles between the two experimental manipulations. Escalating the N-back level in N-LEVELS was associated with greater overall task demand.

Physiological analyses encompassed spectral bandpower features, composite mental workload indices, and event-related potential markers, specifically the N2 and P3b components. Exploratory locally corrected spectral analyses indicated that N-LEVELS produced more frequent modulation of beta-band activity and composite engagement indices. Conversely, N-SPEED was characterized by more frequent posterior and parietal alpha suppression along with theta- and alpha-related power modulation. Event-related potential analyses demonstrated reduced P3b amplitude under both higher N-back levels and accelerated stimulus pacing.

Linear classification models were evaluated using within-subject training and testing alongside leave-one-subject-out validation protocols, incorporating analyses of observation duration, time-on-task dynamics, temporal generalization, cross-paradigm transfer, and exploratory ordinal regression. Decoding across the three conditions of N-SPEED proved more consistent than decoding in N-LEVELS. All three pairwise contrasts within N-SPEED survived statistical correction under both within-subject and leave-one-subject-out cross-validation, whereas only the 1-back versus 3-back contrast survived correction under leave-one-subject-out evaluation in N-LEVELS. Exploratory ordinal regression further indicated a clearer ordered relationship across N-SPEED conditions than across N-LEVELS.

Artifact-sensitivity analyses affirmed the principal P3b findings and largely preserved reference-decoding performance and alpha time-on-task dynamics. However, removing the delta frequency band caused both three-class leave-one-subject-out results to lose significance, and neither the slow-versus-medium nor the medium-versus-fast leave-one-subject-out contrast in N-SPEED survived correction; the slow-versus-medium contrast under leave-one-subject-out testing also failed to survive either high-pass filter variant. Time-on-task tracking revealed selective non-stationarity marked by progressive alpha build-up, while cross-paradigm transfer failed to survive corrected statistical inference. The authors noted that small sample size, deterministic block ordering, modest classification accuracy, and limited cross-paradigm transfer constrain causal attribution and broader applicability beyond the present experimental setup.

The research stems from an international engineering collaboration between academic and industrial neurotechnology teams, bringing together the BioEngineering Lab at the University “G. d’Annunzio” of Chieti-Pescara in Italy and g.tec medical engineering GmbH in Schiedlberg, Austria, with correspondence directed by Arcangelo Merla. Published in the cognitive neuroergonomics section of Frontiers in Neuroergonomics, the investigation addresses a primary challenge in passive brain-computer interface development: determining whether low-density electroencephalography setups can reliably track mental workload across fluctuating task environments, varying observation windows, and different individual operators.

Subject of Research: Electroencephalography monitoring of mental workload during N-back tasks

Article Title: Condition-associated EEG signatures in N-back tasks: temporal dynamics, classification, and generalization

Article References: Spaltro, G., Serioli, L., Sieghartsleitner, S., Gruenwald, J., Guger, C., & Merla, A. (2026). Condition-associated EEG signatures in N-back tasks: temporal dynamics, classification, and generalization. Frontiers in Neuroergonomics, 7, 1927735. https://doi.org/10.3389/fnrgo.2026.1927735

Image Credits: AI Generated

DOI: 10.3389/fnrgo.2026.1927735

Keywords: electroencephalography, mental workload, N-back task, cognitive neuroergonomics, brain-computer interface, machine learning classification, event-related potentials, temporal dynamics, Condition-associated, signatures, N-back, tasks