Predicting and phase targeting brain oscillations in real-time

Objective: Closed-loop neurostimulation (CLNS) procedures, aligning stimuli with electrical brain activity, are quickly gaining popularity in neuroscience. They have been employed to reveal causal links between neural activity patterns and function, and to explore therapeutic effects of electroencephalography (EEG-)guided stimulations during sleep. Most CLNS procedures are developed for a single purpose, detecting one specific pattern of interest in the EEG. Furthermore, most procedures have limited, if any, flexibility to adapt to temporal or interindividual variance in the signal, which means they wouldn't work optimally across the full physiological phenomenology. Approach: Here we present a new approach to CLNS, based on real-time signal modelling to predict brain activity, allowing targeting of a broad variety of oscillatory dynamics. Intrinsic to the modelling approach is adaptation to signal variance, such that no personalization steps prior to use are necessary. We systematically assess stimulus targeting performance of modelling-based CLNS (M-CLNS), across a wide range of brain oscillation frequencies and phases in human and rodent neurophysiological signals. Main results: Our results show high performance for all target phases and frequency bands, including slow oscillations, theta and alpha waves. Significance: These findings highlight the general applicability and adaptability of M-CLNS, which also favors its application in populations with an atypical oscillatory signature, like clinical or elderly populations. In conclusion, M-CLNS provides a promising new tool for neural activity-dependent stimulation in both experimental research and therapeutic applications, such as enhancing deep sleep in patients with sleep disorders.