Motor Imagery BCI Rehabilitation After Stroke

This study will be carried out as an explorative study in order to investigate the benefits of mental imagery training of hand movements guided through Brain-Computer Interface (BCI) feedback after stroke. Stroke patients will be trained to mentally imagine the opening and closing of the hand (hereafter named MI, Motor Imagery). During the training, the patients will receive visual feedback in real time that reflects the neural activity related to motor processes. The NeuroFeedback (NF) will be projected with minimal time delay to maximize the neural learning. This type of brain training with feedback is thought to have significant importance to stimulate the ability of the brain to reorganize and compensate for a damaged region.

Each participant will go through the following data collection procedure (total of 33 measurement sessions per RP):

Week 1-3:

Clinical baseline evaluations, 1 time/week MRI measurements, 1 time/week during 2 weeks. Baseline EEG recordings, 1 time/week Week 4-7 MI-neurofeedback training 3 times/week Clinical intervention evaluation, 1 time/week Week 8-10 Same as week 1-3

Magnetic Resonance Imaging (MRI) measurements. The MRI exam will be carried out on a Siemens MAGNETOM Prisma 3T scanner (head-coil with 20 channels) at baseline and at final assessment session at Stockholm University Brain Imaging Centre. The MRI protocol comprises i) anatomical whole brain spin-echo T1 and T2 weighted sequences for description of lesion size and location ii) acquisition of T2*-weighted gradient echo EPI-BOLD images of the whole brain for assessment of resting state functional connectivity of sensorimotor networks (resting-state functional MRI (fMRI)), and iii) the same sequence as the previous with rest interleaved by a motor imagery paradigm further described below.

Motor Imagery (MI) paradigm. The paradigm consists of instructing RP, by the use of a mirrored computer screen, to either i) rest his/her mind with eyes closed, ii) mentally imagine a hand movement (MI), or ii) execute a hand movement. The hand movements that are instructed are either to close the hand or to open the hand and extend the fingers. RP will perform several repetitions of each hand movement (MI and execution) in order to collect a statistical basis.

Baseline EEG recording. During the baseline EEG recordings, RP will be seated in front a computer screen and perform the MI paradigm (described above). During these session, EEG, EOG, EMG, and accelerometer-data will be collected and are further described below.

ElectroEncephaloGram (EEG), ElectroOculoGram (EOG), ElectroMyoGram (EMG) and accelerometer equipment. The EEG equipment consists of a 64-electrode scalp EEG acquisition system (Brain Products ActiCHamp). The 64 electrodes (active Ag/AgCl) will be distributed according to the extended 10-20 reference placement system. In addition to the EEG recording, 3 electrodes (passive Ag/AgCl, Brain Products) will be placed on each side of both eyes and on the earlob to measure eye-movements during the experiment (EOG). EMG electrodes (passive Ag/AgCl, Brain Products) will be placed over four muscles controlling the wrist and fingers according to a standardized protocol. Two accelerometer-sensors (Brain Products) will be placed on the hand and the index finger in order to record movement-related activity.

EEG, EOG, EMG and accelerometer data analysis. The recorded data will be further analyzed offline in order to evaluate the characteristic features in the data that best describe MI of hand movements. This will be performed in Matlab and Labview combining custom-made scripts with already developed toolboxes (such as EEGLab, Chronux). Features to be evaluated will include the evoked activity, the time-frequency spectra, phase, correlation coefficients, coherency among other. When the feature that best describes MI has been identified different classifier and pattern recognition methods will be evaluated in extracting the information. Intelligent algorithms, Support Vector Machine (SVM), regularized linear regression, naïve Bayes classifiers among others will be evaluated and compared. These are commonly used methods in the field of neurotechnology and a prior comparison-study using neural data from invasive recordings shows the importance of choosing a well-adapted classifier for extracting information.

MI-NeuroFeedback Training (NFT). EEG, EOG, EMG and accelerometer-data will be collected as described in the section "EEG, EMG and accelerometer equipment". RP will perform the MI paradigm without the execution of hand movements. Real-time feedback from recorded EEG-activity will be provided to RP during MI. The feedback consists of a virtual hand on a computer screen whose movements reflect the brain activity of RP related to MI. The recorded data will be further analyzed offline with the analytic tools that are described in previous section.