Skip to main content
Fig. 4 | Journal of NeuroEngineering and Rehabilitation

Fig. 4

From: Semi-automated approaches to optimize deep brain stimulation parameters in Parkinson’s disease

Fig. 4

GP fits after Brute-force and BayesOpt optimization programming. Two PD participants (column) have had their rigidity response curve measured using two different sampling methods. Brute-force method tested all frequencies spaced by 5Hz from 10-185Hz, and sampled in pseudorandom order, top row. Bayesian optimization guided frequency testing for rigidity, bottom rows. Smaller plots shows the fitted GP model evaluated at 3, 6, and 9 frequencies selected by the BayesOpt method. The bottom row shows the final GP fitted model after 12 iterations of the BayesOpt method. Black dashed lines indicate the participant’s RoMaR value after 1 h off stimulation. Red dot indicates the estimated frequency that minimizes rigidity. Shaded gray region indicates RoMaR values within \(\mu \pm \sigma\) of the optimal frequency and the frequency range is defined as all frequencies whose estimated RoMaR values fall within this range and is indicated by the two thin vertical lines

Back to article page