EEG-guided robotic control of transcranial magnetic stimulation

Abstract

Brain-state-guided and closed-loop transcranial magnetic stimulation (TMS) protocols have emerged as methods for decreasing the outcome variability and increasing the therapeutic effectiveness of stimulation protocols. However, existing brain-state-dependent TMS systems control only stimulation timing, while the location is fixed and manually adjusted between blocks or sessions. This limits targeting distributed cortical networks in a flexible brain-state-dependent paradigm. We developed a method that jointly manages TMS pulse timing and location automatically controlled by electroencephalography (EEG)-derived brain state. A machine-learning algorithm infers the brain state in real time to guide robotic coil placement. We present a proof-of-concept study in which EEG controlled both the target site and the timing of TMS. A pre-trained convolutional neural network discriminated between resting state and movements performed with the right or left hand; the classifier output determined the hemisphere in which primary-motor-cortex hand area was stimulated. Preprocessing and decoding of 2-s EEG segments required 150 ms, and the robot took 7.5 s to move from the vertex home position to the predefined motor targets. The EEG-guided robotic TMS method expands the toolkit for brain-state-dependent neurostimulation by enabling closed-loop control of stimulus location based on EEG-derived brain states, including volitional activity. Thus, the method can benefit both neuroscience research and clinical neuromodulation applications, such as by optimizing stimulation parameters to the brain state that produces optimal rehabilitation outcomes in spinal cord injury or motor disorders.