Batch Bayesian optimization design for optimizing a neurostimulator

The paper introduces an adaptive design for efficiently optimizing the programming of a neurostimulator. This is based on patient-reported preferences and Bayesian optimization techniques. The device is programmed with configurations, and patient preferences are recorded. These preferences are then used to update the configuration for the next follow-up period. The process balances exploration of different device settings with maximizing the patient's reported preferences, repeating until a stopping rule is met or the calibration period ends.

• 1
  Batch Bayesian optimization can be used to efficiently identify device configurations for long-term use with near-best patient preference.
• 2
  Early stopping rules can reduce the duration of the calibration phase without significantly compromising the quality of the selected configuration.
• 3
  The q-EI acquisition strategy tended to assign higher-quality configurations throughout the calibration.