Related Experiment Video
Updated: Jul 5, 2026
06:34
A Single-Channel and Non-Invasive Wearable Brain-Computer Interface for Industry and Healthcare
Published on: July 7, 2023
Overcoming brain non-stationarity: adaptive RLS classification for stable BCIs based on auditory evoked potentials
Dovilė Kurmanavičiūtė1, Matilda Makkonen1, Ivan Zubarev1
1Department of Neuroscience and Biomedical Engineering, Aalto University, P.O. Box 12200, FI-00076 Aalto, Finland.
Journal of Neural Engineering
|July 3, 2026
Summary
Adaptive recalibration significantly improves brain-computer interface (BCI) performance for auditory attention decoding, overcoming brain-state non-stationarity. This adaptive approach enhances real-time communication restoration for individuals with disabilities.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Brain-computer interfaces (BCIs) aim to restore communication by decoding selective attention from auditory evoked potentials.
- Clinical translation of auditory BCIs is challenged by brain-state non-stationarity, which degrades decoding performance.
Purpose of the Study:
- To evaluate the impact of different classification strategies on auditory BCI performance under various evaluation settings.
- To assess the effectiveness of adaptive recalibration in mitigating performance decline due to brain-state non-stationarity.
Main Methods:
- Compared offline, causal, static real-time, and adaptive real-time (Recursive Least Squares - RLS) classifiers using 62-channel EEG data from 25 healthy adults.
- Evaluated classifiers in offline (5-fold cross-validation), causal (20/80 split), simulated real-time (static and adaptive), and leave-one-subject-out (LOSO) settings.
Main Results:
- Offline classifiers achieved a peak ROC AUC of 0.75. Causal and static real-time classifiers showed performance degradation (ROC AUC 0.63 and 0.51, respectively).
- Adaptive RLS classifier significantly improved real-time performance, achieving ROC AUC 0.68 and ITR 1.42 bits/min (p < 0.001).
- LOSO evaluation demonstrated RLS feasibility for zero-calibration deployment with trial-by-trial personalization (ROC AUC 0.57, ITR 0.86 bits/min).
Conclusions:
- Brain-state non-stationarity is a primary factor limiting auditory BCI performance.
- Lightweight adaptive recalibration substantially restores real-time decoding performance.
- Adaptive strategies support the translational potential of ERP-based communication paradigms for restoring communication.
