Decoding Motor Imagery From Non-invasive Brain Recordings as a Prerequisite for Innovative Motor Rehabilitation Therapies

Seminal studies in motor neuroscience involving healthy subjects have, since a long time, revealed time-locked changes in induced power within specific frequency band. Brain recordings were shown to exhibit a gradual reduction in signal power, relative to baseline, in the mu (~ 8-12 Hz) and beta (~ 13-30 Hz) frequency bands during an action or during motor imagery (MI): the so-called event-related desynchronization (ERD). This phenomenon is considered to reflect processes related to movement preparation and execution, and is particularly pronounced in the contralateral sensorimotor cortex. Moreover, shortly following the completion of the task a relative increase in power, the event-related synchronization (ERS) (also referred to as the beta rebound), could be observed in the beta band. ERS is thought to reflect the re-establishment of inhibition in the same area.

Ever since the characterization of the ERD and ERS phenomena, there has been little to no discussion in the non-invasive BCI field as to whether these features accurately capture the task-related modulations of brain activity. Recent studies in neurophysiology have challenged this view and have demonstrated that the ERD and ERS patterns only emerge as a result of averaging signal power over multiple trials. On a single trial level, beta band activity occurs in short, transient events, termed bursts, rather than as sustained oscillations. This indicates that the ERD and ERS patterns reflect accumulated, time-varying changes in the burst probability during each trial. Thus, beta bursts may carry more behaviorally relevant information than averaged beta band power. Indeed, studies in humans involving arm movements have established a link between the timing of sensorimotor beta bursts and response times prior to movement, as well as behavioral errors post-movement. Beta burst activity in frontal areas has also been shown to correlate with movement cancellation and recent studies show that activity at the motor unit level also occurs in a transient manner, which is time-locked to sensorimotor beta bursts.

Although beta burst rate has been shown to carry significant information, it still comprises a rather simplistic representation of the underlying activity. Every burst can be characterized by a set of TF-based features: the burst peak time and peak frequency, as well as its duration and its span in the frequency axis. In turn, all these descriptors are extracted using a particular time-frequency transformation and constitute simpler representations of the more complex burst waveform that is embedded in the raw signals, and which is characterized by a stereotypical average shape with large variability around it. The waveform features are neglected in standard BCI approaches, because conventional signal processing methods generally presuppose sustained, oscillatory and stationary signals, and are thus inherently unsuitable for analyzing transient activity.

In contrast to beta, activity in the mu frequency band is oscillatory even in single trials. This activity is typically analyzed using time-frequency decomposition techniques, which assume that the underlying signal is sinusoidal. However, there is now growing consensus that oscillatory neural activity is often non-sinusoidal, and that the raw waveform shape can be informative of movement. Future efforts could take advantage of this possibility by using recently developed non-parametric cycle-by-cycle analyzes.

In this project, the investigators will optimize the design of a subject-specific neurophysiological model to guide motor BCI training, by using Magnetic Resonance Imaging (MRI) and Magnetoencephalography (MEG) for high spatial and biophysical specificity in the experimental group. Anatomical MR volumes will be used to design and 3D-print an individual head-cast that will be used in the MEG scanner in order to stabilize the subject's head position and minimize movements. This high precision approach (hpMEG) has been proven to significantly improve source localization up to the level of distinguishing laminar activity, which makes it a superior-to-EEG recording technique. In MODECO the investigators will use an individualized hpMEG approach, as well as the widely adopted EEG, to study bursts of oscillatory activity in the beta and mu frequency bands related to motor imagery and motor execution. hpMEG will yield subject-specific models of motor imagery that will be used to constrain online decoding of EEG data. This approach will be applied and validated on a group of healthy adult subjects (control group) and will then be compared against another feasibility group of patients (patient group) and age-matched healthy participants (control group; the investigators will attempt to recruit patients relatives). the investigators will compare the proposed approach with a classic EEG-based BCI approach.

the investigators will investigate how to use this information to optimally guide subsequent EEG-based BCI training in the control group. After a thorough investigation in healthy subjects in this project, the investigators will be able to evaluate this approach on a population of stroke patients with upper-limb motor deficits. Two tasks have been designed in this project tasks 1.1 and task 1.2. The aim of Task 1.1 is to develop subject-specific generative models decoding movement onset and /offset, the type of movement (left versus right), as well as finely discretized movement amplitude during both real and imagined wrist extension/flexion movements. In task 1.2 the investigators aim to investigate how the lesions of patients alter our ability to decode attempted wrist movements (control vs patients group).