Fig. 3

The process of real-time closed-loop DBS programming. The recording unit records the biomarker signal via an inserted electrode inside (I) or outside (II) of the brain based on the biomarker type. After signal conditioning (amplification and filtration), the biomarker signal is digitized and then sent to the controller unit. Then, through a computational model (A), the biomarker signal is evaluated from different aspects (e.g. amplitude, frequency, and pulse-width, etc.) to define the response signal, which is then employed to predict optimized stimulation parameters. The bottom model (B) has been adopted from [105] and then modified. It represents the structure of controller where X (t), Y (t), and Z (t) are the input vector, neural circuit states, and biomarker response, respectively. The mapping functions from input to the neural state and from neural state to the biomarker response are demonstrated by f (X,t) and g (X,Y,t), respectively. The k (Z,t) is the controller that evaluates the biomarker response and updates stimulation parameters. The estimated parameters are adjusted in the stimulation unit to create the stimulation pulses for applying to the stimulation electrodes. In this real-time process, a very short time-window of the recorded signal is used for prediction of the stimulation parameters. The time-window of biomarker signal is pushed forward continuously and simultaneous computations are done to predict and update the next stimulation-window