Patient specific Parkinson's disease detection for adaptive deep brain stimulation

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Continuous deep brain stimulation (DBS) for Parkinson's disease (PD) presents challenges including side effects and reduced battery longevity.
• Adaptive stimulation strategies are needed to overcome these limitations, requiring precise, patient-specific PD detection.

Purpose of the Study:

• To develop a patient-customized detector for Parkinson's disease (PD) to enable adaptive deep brain stimulation (DBS).
• To address the variability in biomarkers across patients and time by creating individualized detection models.

Main Methods:

• Utilized patient-specific feature extraction based on spectral band ratios (PD vs. non-PD).
• Employed adaptive Support Vector Machine (SVM) classifiers that adjust decision boundaries for personalized PD detection.
• Developed individualized feature and classifier pairs for each patient.

Main Results:

• Achieved high classification accuracy, exceeding 98% in six out of nine patient datasets.
• Demonstrated the effectiveness of patient-specific feature extraction and adaptive SVM for PD detection.
• Validated the method using local field potential datasets from PD patients.

Conclusions:

• The proposed adaptive detector offers a promising solution for patient-customized PD detection.
• This approach is suitable for on-chip implementation, paving the way for practical adaptive DBS systems.
• Individualized detection can significantly improve the efficacy and reduce the drawbacks of deep brain stimulation in Parkinson's disease treatment.