SAFE-OPT: a Bayesian optimization algorithm for learning optimal deep brain stimulation parameters with safety...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Deep brain stimulation (DBS) requires lengthy, manual parameter tuning, delaying effective treatment for neurological and psychiatric disorders.
• Conventional Bayesian optimization (BO) offers efficient parameter searching but lacks patient-specific safety constraints, risking adverse side effects.

Purpose of the Study:

• To develop and validate SAFE-OPT, a Bayesian optimization algorithm incorporating subject-specific safety constraints for DBS parameter selection.
• To ensure patient safety by avoiding potentially harmful stimulation settings during the optimization process.

Main Methods:

• Developed SAFE-OPT, a Bayesian optimization algorithm designed to learn and enforce subject-specific safety constraints.
• Validated SAFE-OPT using a rodent model of spatial memory deficits induced by multielectrode stimulation.
• Simulated SAFE-OPT configurations *in silico* to optimize its performance for safe and efficient searching.

Main Results:

• SAFE-OPT successfully identified optimal high stimulation amplitudes without compromising task performance in rodents.
• The algorithm demonstrated comparable sample efficiency to conventional BO while strictly adhering to safety thresholds.
• SAFE-OPT avoided selecting stimulation amplitudes that exceeded individual subject safety limits.

Conclusions:

• Integrating subject-specific safety constraints into Bayesian optimization, as demonstrated by SAFE-OPT, is crucial for safe and effective DBS parameter tuning.
• This approach represents a significant step towards the clinical adoption of automated, safe optimization for deep brain stimulation therapies.