Neural mechanisms of training in Brain-Computer Interface: A Biophysical modeling approach

Abstract

Brain-Computer Interface (BCI) is a system that translates neural activity into commands, allowing direct communication between brain and external devices. Despite its clinical application, BCIs are unable to robustly capture the intent of the subjects due to limited understanding of the neural mechanisms underlying BCI control. To address this, we introduce a biophysical linear neural mass model to investigate the associated neural mechanisms of motor imagery BCI (MI-BCI) by modulating the scale-free spectral characteristics of afferent inputs. We tailor this motor imagery neural mass model (mi-NMM) to simulate both MI and resting-state activity and apply it to a cohort of 19 healthy subjects trained over four sessions, during which magnetoencephalography and electroencephalography signals were simultaneously recorded. Task-dependent significant modulation of scale-free cortical dynamics reveals training-induced excitatory-inhibitory (E/I) reconfiguration during BCI training, shifting from widespread cortical to targeted sensorimotor recruitment. The modeled intra-regional E/I interaction strengths and excitatory population time constants capture condition and session-related changes in neural activation in the relevant areas of the sensorimotor cortex, where MI-related desynchronization and synchronization are associated with excitatory population time constants. Thus, these mechanistic findings provide new insights into the underlying neural mechanisms of MI-BCI, paving the way for tailored BCI training protocols.