Automated Sleep Detection in Movement Disorders Using Deep Brain Stimulation and Machine Learning

Area of Science:

• Neuroscience
• Biomedical Engineering
• Sleep Medicine

Background:

• Automated sleep detection in movement disorders is crucial for monitoring and potentially guiding adaptive deep brain stimulation (DBS).
• Understanding wake-versus-sleep status (WSS) using local field potentials (LFP) is essential for patients with movement disorders.

Purpose of the Study:

• To compare local field potentials (LFP) during wake versus sleep status (WSS) in a home environment.
• To develop biomarkers for WSS in patients with Parkinson's disease (PD), essential tremor (ET), and Tourette's syndrome (TS).

Main Methods:

• Recorded β-band and/or α-band LFP power spectral densities over 5-7 nights in PD, ET, and TS patients implanted with Medtronic Percept devices.
• Utilized wearable actigraphy to track sleep patterns concurrently with LFP recordings.
• Trained machine learning classifiers to differentiate between wake and sleep states based on LFP data.

Main Results:

• Parkinson's disease (PD) patients showed increased LFP β-power in the subthalamic nucleus (STN) and decreased power in the globus pallidus interna (GPi) from sleep to wake; α-power increased in both.
• Machine learning models achieved high WSS classification accuracy: 93% for PD, 86% across all patients, 86% for ET, and 89% for TS.
• Chronic intracranial narrowband recordings demonstrated accurate sleep identification in diverse movement disorders.

Conclusions:

• Chronic intracranial narrowband recordings offer a viable method for accurately identifying sleep states in patients with various movement disorders.
• This proof-of-concept study highlights the potential of LFP biomarkers for WSS monitoring in PD, ET, and TS.
• The findings support the development of adaptive deep brain stimulation (DBS) strategies guided by real-time sleep detection.