Customized Intracranial Sleep Stage Decoding During Deep Brain Stimulation in Parkinson’s Disease

Overview

This study demonstrates that personalized machine learning models can accurately classify sleep stages from intracranial recordings in Parkinson’s disease patients undergoing deep brain stimulation. Both supervised and unsupervised approaches achieved high accuracy, supporting the feasibility of adaptive DBS targeting sleep disruptions.

Background

Sleep disturbances are a prevalent and challenging non-motor symptom in Parkinson’s disease, significantly impacting quality of life. Adaptive deep brain stimulation (aDBS) that targets specific sleep stages may offer therapeutic benefits. However, reliable real-time sleep stage classification from intracranial signals during stimulation is required to enable such interventions. This study evaluates machine learning methods for decoding sleep stages from multi-night intracranial cortico-basal recordings synchronized with scalp EEG in PD patients.

Data Highlights

Key Findings

• Multi-night intracranial recordings from 5 PD patients undergoing chronic DBS were used to classify sleep stages.
• Supervised machine learning models achieved an average five-stage sleep classification accuracy of 80.2% (±0.9% SEM).
• Binary classification of NREM sleep using linear models compatible with current DBS devices reached 85.9% accuracy (±0.4% SEM).
• Unsupervised clustering combined with linear models yielded 83.5% accuracy (±5.6% SEM) for NREM sleep discrimination.
• These results confirm the feasibility of personalized sleep decoding during DBS using intracranial signals.

Clinical Implications

The ability to accurately decode sleep stages intracranially during DBS paves the way for adaptive stimulation protocols tailored to sleep architecture in Parkinson’s disease. Implementing device-compatible linear models for binary NREM classification could facilitate real-time sleep monitoring and targeted therapy. This approach may ultimately improve sleep quality and overall disease management in PD patients.

Conclusion

Personalized supervised and unsupervised machine learning models can effectively classify sleep stages from intracranial recordings during deep brain stimulation in Parkinson’s disease. These findings support the development of adaptive DBS strategies targeting sleep dysfunction.

References

1. Chaudhuri et al. 2006 -- Non-motor symptoms of Parkinson’s disease: diagnosis and management
2. Diederich et al. 2005 -- Progressive sleep ‘destructuring’ in Parkinson’s disease
3. Weintraub et al. 2022 -- The neuropsychiatry of Parkinson’s disease: advances and challenges
4. Berry et al. 2020 -- The AASM Manual for the Scoring of Sleep and Associated Events
5. Zahed et al. 2021 -- The neurophysiology of sleep in Parkinson’s disease
6. Schreiner et al. 2019 -- Slow-wave sleep and motor progression in Parkinson's disease
7. Chen et al. 2024 -- Correlation of slow-wave sleep with motor and nonmotor progression in Parkinson’s disease
8. Zuzuárregui & Ostrem 2020 -- The impact of deep brain stimulation on sleep in Parkinson’s disease: an update