To analyze the efficacy of personalized supervised and unsupervised machine learning approaches in classifying sleep stages from participants receiving deep brain stimulation (DBS) for Parkinson's Disease.
Key Findings:
Five-stage classification accuracy averaged 80.2% (±0.9% SEM) across PD subjects, indicating a robust performance.
Binary NREM classification using linear models achieved an average accuracy of 85.9% (±0.4% SEM), demonstrating the effectiveness of linear models in practical applications.
Linear models trained on unsupervised cluster labels achieved an average accuracy of 83.5% (±5.6% SEM) for NREM sleep, suggesting potential for unsupervised learning in this context.
Interpretation:
The study demonstrates the feasibility of personalized machine learning models for effective sleep classification using intracranial data during deep brain stimulation.
Limitations:
Datasets are not publicly available due to personal health information restrictions, which limits reproducibility.
Underlying code and training datasets may not be accessible without reasonable request, potentially hindering further research.
Conclusion:
Personalized supervised and unsupervised machine learning approaches show promise for improving sleep stage classification in Parkinson's Disease patients undergoing DBS, paving the way for future research and clinical applications.
