RL-roboTMS

DEEP REINFORCEMENT LEARNING FOR OPTIMIZING ROBOT-ASSISTED CLOSED-LOOP TRANSCRANIAL MAGNETIC STIMULATION

Summary

Transcranial magnetic stimulation (TMS) is a non-invasive brain stimulation tool that is widely usedin clinical practice and in neuroscience research. Although TMS-based therapies for variouspatient groups (e.g., patients with pharmaco-resistant depression and obsessive-compulsivedisorder) have been approved by the US Food and Drug Administration (FDA), their ability toreduce disabilities in a functionally significant and sustained manner remain scarce. In workleading to this proposal, we used multi-scale computational modeling and demonstrated a highvariability in TMS-induced electric fields and the expected biological effects when currentlyestablished FDA-approved intensity selection approaches are employed. We posit that part of theresponse variability to TMS interventions-and hence therapeutic success-stems from the currentmethods of parameter selection that do not address inter- and intraindividual variabilities in brainanatomy and activity.

The proposed project aims at establishing an algorithmic pipeline of deep reinforcementlearning for optimized robot-assisted closed-loop TMS, thus paving the way towards personalizedbrain stimulation in neuroscience research and therapeutic settings. We will initially focus on themotor cortex of healthy human subjects, because it is by far the most frequently studied corticalarea in the human brain, and second, because stimulating this cortical region can produce motorevoked potentials (MEPs) that can be readily used to quantify the effect of TMS (i.e., ideal forclosed-loop applications).