“…Therefore, ML can play a guiding role in online adaptive stimulation. 365 , 366 , 367 For instance, the feedback loop can analyze the neural signal’s oscillatory patterns or other reliably detectable biosignals (e.g., biochemcial, electromyographic, and mechanical signals) to classify or detect the critical brain state for delivery of closed-loop neurostimulation. 368 Additionally, reinforcement learning can be applied to learn a state-action value function to identify the best excitability brain state, where the state corresponds to the neural activity (e.g., the amplitude of evoked potentials, characteristics of brain connectivity) and the action corresponds to on/off stimulation mode.…”