Decoding of Turning Intention during Walking Based on EEG Biomarkers.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Existing brain-computer interface (BCI) literature lacks asynchronous intention models for real-time applications.
• Electroencephalography (EEG) has not been fully validated as a real-time biomarker for specific intentions like turning.

Purpose of the Study:

• To propose and validate a novel BCI approach for real-time detection of the intention to turn.
• To develop a BCI methodology capable of differentiating between monotonous walking and the intention to turn.
• To compare the efficacy of popular signal processing and classification algorithms for this BCI application.

Main Methods:

• Utilized electroencephalography (EEG) signals to develop a BCI for intention detection.
• Employed H∞ filter and ASR for signal filtering, and Riemannian space analysis for processing and classification.
• Compared various classifiers to estimate the distance of test samples in Riemannian space, avoiding power-based models and baseline correction.

Main Results:

• Achieved average cross-validation accuracies of 66.2% (generic) and 69.6% (personalized) using the H∞ filter.
• In pseudo-online tests, the best subject achieved 43.9% True Positives (TP) and 2.9 False Positives per minute (FP/min).
• The final validation demonstrated 71.43% accuracy with 2.5 FP/min and 0.21s turn anticipation for the best subject.

Conclusions:

• EEG can serve as a reliable biomarker for detecting the intention to turn in real-time.
• The proposed Riemannian space-based BCI methodology effectively differentiates walking from turning intentions.
• This BCI approach shows significant potential for real-time applications, improving upon existing limitations.