Real-time EEG feedback during simultaneous EEG–fMRI identifies the cortical signature of motor imagery

Highlights

• Real-time EEG feedback during continuous fMRI is possible for motor imagery.

• EEG feedback enhances task-related activity in EEG and the feedback independent fMRI.

• BCI illiterates can be divided into real and pseudo BCI illiterates.

Abstract

Motor imagery (MI) combined with real-time electroencephalogram (EEG) feedback is a popular approach for steering brain–computer interfaces (BCI). MI BCI has been considered promising as add-on therapy to support motor recovery after stroke. Yet whether EEG neurofeedback indeed targets specific sensorimotor activation patterns cannot be unambiguously inferred from EEG alone. We combined MI EEG neurofeedback with concurrent and continuous functional magnetic resonance imaging (fMRI) to characterize the relationship between MI EEG neurofeedback and activation in cortical sensorimotor areas. EEG signals were corrected online from interfering MRI gradient and ballistocardiogram artifacts, enabling the delivery of real-time EEG feedback. Significantly enhanced task-specific brain activity during feedback compared to no feedback blocks was present in EEG and fMRI. Moreover, the contralateral MI related decrease in EEG sensorimotor rhythm amplitude correlated inversely with fMRI activation in the contralateral sensorimotor areas, whereas a lateralized fMRI pattern did not necessarily go along with a lateralized EEG pattern. Together, the findings indicate a complex relationship between MI EEG signals and sensorimotor cortical activity, whereby both are similarly modulated by EEG neurofeedback. This finding supports the potential of MI EEG neurofeedback for motor rehabilitation and helps to better understand individual differences in MI BCI performance.

Introduction

On average, every 40 s someone suffers a stroke in the United States alone (Roger et al., 2012). At present, physical therapy is the preferred treatment to recover from stroke induced movement impairments, aiming to reorganize neural networks and restore motor functions (Teo and Chew, 2014). However, stroke remains the leading cause of long-term disability worldwide (Ward and Cohen, 2004). Functional neuroimaging studies have revealed that, after stroke, changes can take place resulting in a widespread functional reorganization of the motor network, even at cortical areas distant from a focal lesion (Ward, 2011). Specifically, regarding attempted voluntary movements of the paretic hand a relatively strong contralesional cortical activation alongside a relatively weak perilesional activation has been identified as potentially detrimental for motor recovery. Restoring the originally stronger cortical activation of the hemisphere contralateral to the paretic hand may be crucially beneficial (Grefkes and Ward, 2014). Among the interventions facilitating such a restoration, motor imagery (MI), a mental practice strategy, has been proposed as a complement to traditional rehabilitation approaches (Buch et al., 2008). MI is the mental rehearsal of a particular motor act without any overt motor output. According to the neural simulation of action account, movement imagination partly activates the same motor network that is responsible for motor execution (Jeannerod, 2001). Consequently, MI may open a door towards experience-driven cortical reorganization and functional restoration, even in the absence of residual voluntary limb movement.

To facilitate adaptive recovery of lost motor functions, high-intensity, task-specific practice with feedback on performance is needed (Langhorne et al., 2009). The use of brain–computer interface (BCI) technology allows to decode brain activation patterns generated by MI. The result of the decoding procedure can be displayed to the user, creating a feedback reflecting task performance. Electroencephalogram (EEG) based MI feedback systems are particularly promising for therapeutic applications because mobile EEG systems are available that are small in size, easy to use, and low in cost to allow regular use, even at home, and thus uniquely enable high-intensity practice (De Vos et al., 2014, Debener et al., 2012, Kranczioch et al., 2014, Zich et al., 2015). EEG-based feedback for MI typically reflects the amplitude reduction of brain oscillations from sensorimotor cortical areas (Cheyne, 2013). When imagining or executing movements these sensorimotor rhythms (SMRs) desynchronize in the 8 to 30 Hz frequency range, a pattern of activity known as event-related desynchronization (ERD) (Pfurtscheller and Neuper, 2001). A recent study on healthy adults provides evidence that ERD patterns can be changed by EEG feedback. In detail, participants underwent 4 days of MI training with EEG neurofeedback after which they showed a stronger ERD lateralization (Zich et al., 2015). However, because of the inverse problem inherent to EEG it cannot be unambiguously inferred that task-specific ERD effects induced by MI uniquely relate to a change in sensorimotor activation patterns. One way to address this issue is to combine EEG neurofeedback and fMRI measurements. This has been done before, but with off-line and separate recording study designs (Formaggio et al., 2010, Halder et al., 2011, Yuan et al., 2010), which again suffer from the inverse and interpretational problems.

Aiming to better validate and characterize MI EEG neurofeedback effects a study was designed in which EEG-based feedback was combined with simultaneous and continuous fMRI. This was achieved by correcting the EEG signal from interfering MRI gradient (GA) and ballistocardiogram artifacts (BCG) in near real-time. First efforts towards implementing online EEG-based feedback inside the MRI scanner demonstrate the potential of this approach (Zotev et al., 2014). Here we used a popular BCI paradigm, namely MI, with online EEG feedback and simultaneous fMRI. EEG data were recorded inside the MRI scanner and validated by comparing the ERD characteristics to separate EEG recordings made outside the scanner from the same participants. Subsequently the electrophysiological and hemodynamic signatures of MI EEG feedback effects were characterized. Specifically, we predicted that EEG feedback compared to no feedback enhances task-specific brain activity in both EEG and fMRI modalities. In addition, a systematic correlation between EEG SMR amplitude reduction and fMRI activity in sensorimotor areas was hypothesized. Verification of both predictions would indicate that MI EEG feedback is directly related to sensorimotor cortex activation and that MI EEG feedback can manipulate activation patterns in these areas. Moreover, capitalizing on the complementary advantages of temporally accurate EEG-based feedback and high spatial resolution neuroimaging we identified systematic patterns of inter-individual differences in MI BCI performance. This issue is of considerable practical relevance because the large, unexplained inter-individual variability in MI BCI performance hampers extensive clinical investigations.