Personalized supervised and unsupervised intracranial sleep decoding during deep brain stimulation.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Sleep Medicine

Background:

• Sleep disturbances are a major challenge for Parkinson's Disease (PD) patients.
• Adaptive Deep Brain Stimulation (aDBS) offers a potential therapeutic avenue by targeting sleep neurophysiology.
• Personalized machine learning (ML) approaches are being explored for sleep stage classification.

Purpose of the Study:

• To evaluate the effectiveness of personalized ML models in classifying sleep stages using intracranial recordings in PD patients undergoing deep brain stimulation (DBS).
• To assess the feasibility of implementing these ML models on current DBS devices for potential therapeutic applications.

Main Methods:

• Acquired 283 hours of intracranial cortico-basal recordings and synchronized scalp EEG sleep stage labels from 5 PD participants during chronic DBS.
• Developed and applied personalized supervised and unsupervised ML models for sleep stage classification.
• Evaluated classification accuracy, including models suitable for real-time implementation on DBS devices.

Main Results:

• Average five-stage sleep classification accuracy reached 80.2% across PD subjects.
• Binary NREM sleep classification using linear models, implementable on current DBS devices, achieved 85.9% accuracy.
• Linear models trained on unsupervised clusters for NREM discrimination showed 83.5% accuracy.

Conclusions:

• Personalized supervised and unsupervised ML models are feasible for classifying sleep stages using intracranial data during DBS in PD patients.
• These findings demonstrate the potential for ML-driven adaptive DBS to improve sleep quality in Parkinson's Disease.
• The study highlights the viability of using advanced computational techniques for personalized neurological therapies.