Detection of Error-Related Potentials in Stroke Patients from EEG Using an Artificial Neural Network.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Science

Background:

• Error-related potentials (ErrPs) are neural signals indicating errors, proposed for enhancing brain-computer interface (BCI) performance.
• In stroke rehabilitation, ErrPs could minimize BCI calibration time by enabling continuous adaptation and individualized BCI performance.
• Accurate labeling of ErrP data is crucial for effective BCI adaptation in stroke patients.

Purpose of the Study:

• To classify single-trial ErrPs in individuals with stroke.
• To assess the test-retest reliability of ErrP classification.
• To compare different classifier calibration schemes (within-day, between-day, across-participant) using artificial neural networks (ANN) and linear discriminant analysis (LDA) with waveform features.

Main Methods:

• Twenty-five individuals with stroke performed a sham BCI task with error/correct feedback, while continuous EEG was recorded.
• EEG data were segmented into ErrP and NonErrP epochs and classified using ANN (temporal features or entire epoch) and shrinkage LDA.
• Waveform features from the sensorimotor cortex were used for physiological interpretability.

Main Results:

• Within-day calibration achieved 90% classification accuracy with ANN using the entire epoch, decreasing to 86% (ANN, temporal features) and 69% (LDA).
• Poor test-retest reliability was observed between sessions; other calibration schemes yielded 63-72% accuracy, with LDA performing best.
• No correlation was found between stroke impairment level and classification accuracy.

Conclusions:

• ErrPs can be classified in individuals with stroke, but optimal decoding necessitates user- and session-specific calibration.
• The use of ErrP/NonErrP waveform features allows for physiologically meaningful interpretation of classifier outputs.
• Findings support the potential of ErrPs for continuous data labeling in BCIs for stroke rehabilitation, potentially improving BCI performance.