A Framework for Adapting Deep Brain Stimulation Using Parkinsonian State Estimates.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Deep brain stimulation (DBS) shows promise for Parkinson's disease (PD) but its mechanisms are not fully understood.
• This lack of understanding impedes the development of adaptive DBS (aDBS) systems.
• Advancing aDBS requires deeper insights into PD dynamics, achievable through machine learning.

Purpose of the Study:

• To develop and evaluate a machine learning approach for real-time estimation of PD symptom severity.
• To enable adaptive therapy adjustments for improved PD management.
• To explore the utility of generative and discriminative models in a closed-loop DBS system.

Main Methods:

• Utilized generative (Gaussian mixture model) and discriminative (support vector machine) machine learning models.
• Integrated machine learning state estimates with a fuzzy controller in a critic-actor control framework.
• Applied the system to estimate symptom severity and adjust therapy in a simulated PD environment.

Main Results:

• Both machine learning model configurations accurately estimated PD symptom severity.
• The adaptive system successfully suppressed PD symptoms to a desired state in 7 out of 9 cases.
• Therapeutic effects were achieved rapidly, with most cases settling in under 2 seconds.

Conclusions:

• Machine learning models can effectively estimate PD symptom severity for adaptive DBS.
• The proposed critic-actor approach with machine learning shows potential for effective PD symptom suppression.
• This work contributes to the development of more sophisticated and responsive aDBS systems.