Combining hidden Markov models for comparing the dynamics of multiple sleep electroencephalograms

Area of Science:

• Computational Neuroscience
• Sleep Medicine
• Biostatistics

Background:

• Sleep disorders, particularly sleep-disordered breathing (SDB), have significant cardiovascular consequences.
• Analyzing electroencephalogram (EEG) signals during sleep is crucial for understanding sleep disorders.
• Existing methods for analyzing large-scale, multi-time-series EEG data can be complex and computationally intensive.

Purpose of the Study:

• To develop and apply an easily implemented Hidden Markov Model (HMM) method for analyzing populations of EEG signals during sleep.
• To model multiple time series simultaneously, accommodating large datasets.
• To investigate differences in brain activity during sleep between individuals with and without SDB using HMMs.

Main Methods:

• Utilized the longitudinal cohort from the Sleep Heart Health Study (SHHS) for HMM parameter development.
• Developed a method for simultaneous modeling of multiple EEG time series.
• Calculated subject-specific Markovian predictions and derived indices like transition frequencies between latent states.

Main Results:

• Successfully applied HMMs to analyze EEG data from the SHHS cohort.
• Identified differences in brain activity patterns during sleep between SDB and non-SDB groups.
• Observed stability in the percentage of time spent in HMM latent states across matched groups, but differences in state transition rates.

Conclusions:

• The proposed HMM method provides an effective approach for analyzing large-scale EEG sleep data.
• HMM analysis reveals distinct dynamic brain activity patterns associated with SDB.
• Findings highlight the utility of HMMs in uncovering subtle differences in sleep brain activity relevant to sleep disorders.