Abstract:
Background: In recent years, the interest in the detection of intentions of voluntary movements for triggering external devices in closed-loop brain-computer interfaces (BCIs) for rehabilitation procedures has increased. Lately, for detection, a type of slow cortical potentials (SCPs) called movement-related cortical potentials (MRCPs) has been used.
Objective: The objectives of the study were; to detect the movement intentions from single-trial EEG recordings in real-time; to classify the two types of movements based on different levels of rate of torque development and force; and to evaluate the performance of the system based on the combination of detection and classification.
Method: EEG from 12 healthy subjects performing self-paced real and imaginary isometric dorsiflexions of the ankle and 6 stroke patients performing the actual dorsiflexions was recorded. The template matching technique was used for detection of the movement intentions. Linear support vector machine was used for classification using 5 features extracted from the initial negative phase of MRCPs.
Main Results: The system performance for real and imaginary tasks by healthy subjects was 65.1 ± 2.9 % and 62.1 ± 5.3 %, respectively and that for stroke patients was 50.6 ± 8.6 %. The true positive rates for all types of tasks were more than 88 % and the range for classifier accuracies were 58 – 73 %.
Significance: The results show that the system developed has the potential to detect and classify self-paced movement intentions in real-time. In combination with assistive devices like functional electrical stimulator, this online system can be used for inducing cortical neuroplasticity for neurorehabilation of stroke patients by providing precise afferent feedback in reply to efferent activity.
