Multi-objective data-driven optimization for improving deep brain stimulation in Parkinson's disease.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Science

Background:

• Deep Brain Stimulation (DBS) is a key Parkinson's disease (PD) treatment, but requires extensive trial-and-error for optimal settings.
• Current data-driven algorithms optimize symptom relief but struggle to account for potential side effects.
• A need exists for methods that balance therapeutic benefits with adverse effects in DBS programming.

Purpose of the Study:

• To demonstrate the application of multi-objective data-driven optimization for DBS in Parkinson's disease.
• To identify optimal trade-offs between maximizing symptom relief and minimizing side effects.
• To characterize critical design features for clinical application of these optimization techniques.

Main Methods:

• Developed a framework using cortical and motor evoked potentials from PD patients undergoing subthalamic nucleus DBS.
• Defined two optimization objectives: maximizing cortical evoked potentials (therapeutic biomarker) and minimizing motor evoked potentials (side effect biomarker).
• Employed and evaluated data-driven multi-objective optimization algorithms, including Bayesian optimization with surrogate models.

Main Results:

• The multi-objective optimization framework reduced the stimulation parameter space by 61 ± 19%.
• Bayesian optimization achieved an area under the receiver operating characteristic curve of up to 0.94 ± 0.02.
• Efficient selection of stimulation settings was achieved using surrogate models and tuned acquisition functions.

Conclusions:

• Multi-objective optimization offers a promising strategy for optimizing DBS by balancing symptom relief and side effects.
• This approach is adaptable to new biomarkers, other neurological disorders, and advanced DBS technologies.
• The findings pave the way for more efficient and personalized DBS therapy for Parkinson's disease.