BCIwiki (bciwiki.com) — A multimodal deep learning decoder trained on heart rate, heart rate variability (HRV), respiration, electrodermal activity and pupil diameter decoded task-related arousal with a within-subject AUC of 93.0%, beating an EEG-only decoder at 85.2% and the original filter-bank common spatial pattern (FBCSP) decoder at 79.8%, Anirudh Natarajan and Paul Sajda reported in a preprint posted to arXiv on October 2, 2026 (arXiv:2610.04113). The work has not been peer reviewed.
Most arousal-based brain-computer interfaces (BCIs) decode cognitive state from EEG, but arousal is an autonomic process, the authors noted. They reanalyzed a public dataset from a difficult boundary-avoidance flight task in which participants received EEG-based BCI neurofeedback, sham feedback or no feedback. All analyses were offline.
Only the peripheral decoder reproduced the Yerkes-Dodson inverted-U relation between decoded arousal and performance. Feedback conditions did not change arousal trajectories, which the authors read as a sign that the BCI benefit may come from regulation at critical moments rather than a continuous shift.
The team derived a time-varying optimal arousal trajectory from control trials; per-trial deviations from it predicted performance (r = -0.28 to -0.24, p < 0.01). Baseline HRV and gamma power shaped each subject's arousal-performance curve, and the authors used them to split subjects into arousal-sensitive and arousal-tolerant groups. Group-specific control bands increased trial separation, raising Cohen's d from 1.21 to 1.32 and from 0.85 to 1.10.
The authors say the results offer a specification for individualized closed-loop arousal regulation that a prospective study must test, for example with personalized transcutaneous vagus nerve stimulation.