High Classification Accuracy of a Motor Imagery Based Brain-Computer Interface for Stroke Rehabilitation Training

Abstract

Motor imagery (MI) based brain-computer interfaces (BCI) extract commands in real-time and can be used to control a cursor, a robot or functional electrical stimulation (FES) devices. The control of FES devices is especially interesting for stroke rehabilitation, when a patient can use motor imagery to stimulate specific muscles in real-time. However, damage to motor areas resulting from stroke or other causes might impair control of a motor imagery BCI for rehabilitation. The current work presents a comparative evaluation of the MI-based BCI control accuracy between stroke patients and healthy subjects. Five patients who had a stroke that affected the motor system participated in the current study, and were trained across 10-24 sessions lasting about 1 h each with the recoveriX system. The participants' EEG data were classified while they imagined left or right hand movements, and real-time feedback was provided on a monitor. If the correct imagination was detected, the FES was also activated to move the left or right hand. The grand average mean accuracy was 87.4% for all patients and sessions. All patients were able to achieve at least one session with a maximum accuracy above 96%. Both the mean accuracy and the maximum accuracy were surprisingly high and above results seen with healthy controls in prior studies. Importantly, the study showed that stroke patients can control a MI BCI system with high accuracy relative to healthy persons. This may occur because these patients are highly motivated to participate in a study to improve their motor functions. Participants often reported early in the training of motor improvements and this caused additional motivation. However, it also reflects the efficacy of combining motor imagination, seeing continuous bar feedback, and real hand movement that also activates the tactile and proprioceptive systems. Results also suggested that motor function could improve even if classification accuracy did not, and suggest other new questions to explore in future work. Future studies will also be done with a first-person view 3D avatar to provide improved feedback and thereby increase each patients' sense of engagement.

EEG Biomarkers Related With the Functional State of Stroke Patients

“…For classification, we used linear discriminant analysis (LDA) on the spatially filtered data. We followed the steps described in Irimia et al (2018) except for a change in the electrode setup: FPZ, FP1, FP2, AF7, AF3, AFZ, AF4, AF8, F7, F5, F3, F1, FZ, F2, F4, F6, F8, FT7, FC5, FC3, FC1, FCZ, FC2, FC4, FC6, FT8, T7, C5, C3, C1, CZ, C2, C4, C6, T8, TP7, CP5, CP3, CP1, CPZ, CP2, CP4, CP6, TP8, P7, P5, P3, P1, PZ, P2, P4, P6, P8, PO7, PO3, POZ, PO4, PO8, O1, OZ, O2, O9, O10.…”

Brain-Computer Interface Coupled to a Robotic Hand Orthosis for Stroke Patients’ Neurorehabilitation: A Crossover Feasibility Study

“…These degrees of control are in the same range as those reported by Ang et al (2011) of an average of 74% in 46 stroke patients, using a 27-electrode system. However, in the last session with the ReHand-BCI, patients’ median accuracy was lower than the maximum accuracy range of 97.5–96.2% reported by Irimia et al (2018) in five stroke patients controlling a BCI using 64 electrode positions, but within their minimum reported accuracy range (82.5–60%). A possible reason for this difference could be the enhanced spatial resolution of a BCI system with 64 electrode positions, compared to ReHand-BCI with 11.…”

Brainwave Classification of Task Performed by Stroke Patients using ANN

“…The stroke classification during stroke treatment and recovery has many implications. It works as a reference for prediction, recommendation of treatment, and secondary stroke prevention approaches [3], [4], [5]. A classification is the method of obtaining a number of models that define and separate data classes for the use of unknown class label prediction [6].…”