Functional Interactions of Low‐density Electroencephalography During Sleep Cycles to Detect Alzheimer’s Disease

Abstract

AbstractBackgroundAlzheimer’s disease (AD) causes disruption in the collective dynamics of the brain activities which can be captured by analyzing the functional interactions between electroencephalographic (EEG) signals in different regions of the brain. To better assess the AD‐related changes in the interactions, one possible approach is to increase the EEG electrode density for better spatial information. Alternatively, high order interactions within the fluctuations of low‐density EEG — a concept of the multivariate information theory — may better reveal complex EEG interactions and can be used to detect AD‐related changes in brain activities. This study is aimed to demonstrate this conceptual approach using machine learning.MethodWe analyzed overnight 4‐channel EEG recordings (left central, right central, left mastoid, and right mastoid) in the polysomnography (PSG) data collected from 28 older men in the Osteoporotic Fractures in Men Study (MrOS). 14 participants reported a history of clinical diagnosis of AD while the other 14 did not. Groups were matched in age and education. Each of 30‐second epochs of the PSG recording was scored as wake, N1, N2, N3, or rapid eye movement sleep according to the established guideline. Features based on low (2) and high order (3‐4) interactions between electrodes, i.e., total correlation (TC), dual total correlation (DTC), O‐information (O) and S‐information (S), were extracted from raw and band‐passed filtered data (δ: 0.5‐4 Hz, θ: 4‐8 Hz, α: 8‐12 Hz, β: 12‐30 Hz, γl: 30‐40 Hz, γh: 40‐100 Hz) within each wake/sleep stage. The optimal set of features was identified using a cross‐validated random forest classifier to maximize the group discrimination.ResultAD was discriminated from control with an average area under the receiver operating characteristic curve of 0.90±0.05 using just 3 features: O (triplet) measured in the γh band within wake stage, DTC (triplet) of raw EEG within N1 stage, and TC (pairwise) in the δ band within N1 stage.ConclusionThis proof‐of‐concept study shows the possibility of using low‐density sleep EEG to detect AD. This method might be useful for screening people with risk of AD in ambulatory setting.