An active learning framework for personalized deep brain stimulation

Abstract Background: To personalize deep brain stimulation (DBS), we need to identify a link between DBS parameters and the neural response of an individual. The existing approach based on random or empirical sampling (RS) is time-consuming, costly, practically impossible in clinical setting. This limits discovery novel settings under challenging situations. address this problem, have developed new algorithmic framework active learning (AL) that can obtain best model while minimizing number experiments. Methods: We used computational Parkinson’s disease generate synthetic data. swept subthalamic nucleus amplitude, frequency, pulse width estimating globus pallidus internus (GPi) beta (13-30 Hz) power for each parameter. resulted 200 different samples. randomly selected 80% data pool training reserved remaining 20% as unseen test Using three initial samples, trained two linear regression models (based RS AL) GPi power. iteratively added one sample at time both AL approaches until had 20 then tested iteration calculated root mean squared error (RMSE). process was repeated 1000 times. Results: RMSE 0.043 0.039. Results showed AL-based outperformed RS-based by showing significantly less errors dataset two-sample t-test (p= 2.33e-07, N=1000). Conclusion: validated our outperforms identifying individualized duration experimental procedure. Research Category Technology Methods Translational Research: 1. Deep Brain Stimulation (DBS) Keywords: Personalized Stimulation, Active Learning, Machine Learning