Topographic somatosensory imagery for real-time fMRI Brain-Computer Interfacing

Abstract

Real-time functional magnetic resonance imaging (fMRI) is a promising non-invasive method for brain computer interfaces (BCIs). BCIs translate brain activity into signals that allow communication with the outside world. Visual and motor imagery are often used as information-encoding strategies, but can be challenging if not grounded in recent experience in these modalities, e.g. in patients with locked-in-syndrome (LIS). In contrast, somatosensory imagery might constitute a more suitable information-encoding strategy as somatosensory function is often very robust. Somatosensory imagery has been shown to activate somatotopic cortex, but it has been unclear so far whether it can be reliably detected on a single-trial level and successfully classified according to specific somatosensory imagery content.\n\nUsing ultra-high field 7-T fMRI, we show reliable and high-accuracy single-trial decoding of left-foot vs. right-hand somatosensory imagery. Correspondingly, higher decoding accuracies were associated with greater spatial separation of hand and foot decoding-weight patterns in primary somatosensory cortex (S1). Exploiting these novel neuroscientific insights, we developed - and provide a proof of concept for - basic BCI communication by showing that binary (yes/no) answers encoded by somatosensory imagery can be decoded with high accuracy not only offline but also in real-time.\n\nThis study demonstrates that body part-specific somatosensory imagery differentially activates somatosensory cortex in a topographically specific manner; evidence which was surprisingly still lacking in the literature. It is also offers a promising novel somatosensory imagery based fMRI-BCI control strategy, with particularly high potential for visually and motor-impaired patients. The strategy could also be transferred to lower MRI field strengths and to mobile functional near-infrared spectroscopy. Finally, given that communication BCIs provide the BCI user with a form of feedback based on their brain signals and can thus be considered as a specific form of neurofeedback, and that repeated use of a BCI has been shown to enhance underlying representations, we expect that the current BCI could also offer an interesting new approach for somatosensory rehabilitation training in the context of stroke and phantom limb pain.