Stability of Magnetoencephalography and Electroencephalography Spectral Features in Mild Cognitive Impairment

Abstract

AbstractBackgroundIn mild cognitive impairment (MCI), the first neurodegenerative changes may occur years before they can be effectively detected with current clinical tools. Early pathology has been linked to synaptic dysfunction causing brain network disturbances that may be observed with electroencephalography (EEG) and magnetoencephalography (MEG). These methods are still not widely utilized in dementia risk assessment, and the stability of suggested EEG and MEG features and their correlation has not been thoroughly addressed in this patient group.MethodIn a 5‐year EU Horizon2020 project ‘AI‐Mind’, we collect EEG and simultaneous MEG data from MCI subjects, to be compiled with cognitive, genetic and plasma biomarker measures and techniques based on artificial intelligence (AI) for predicting the overall dementia risk.Here, we analyzed a subset of 31 Finnish MCI participants (20 females, age 68 ± 5; mean ± SD) from the AI‐Mind dataset to assess the stability of prominent signal features from resting‐state EEG and MEG recordings. We determined within‐session test‐retest reliability for spectral peak frequencies and relative peak power over occipital brain regions in the 8‐13 Hz range, by utilizing intraclass correlation (ICC) for subsequent 5‐min runs of EEG and MEG data in both eyes‐closed (EC) and eyes‐open (EO) conditions. In addition, we determined the consistency of the features across methods (EEG/MEG).ResultThe results indicate good (ICC>0.6) test‐retest reliability across runs for EEG (peak frequency: 0.72 (EC); relative power: 0.86 (EC) and 0.66 (EO); p<0.001) and excellent test‐retest reliability (ICC>0.75) for MEG (peak frequency: 0.83 (EC); relative power: 0.96 (EC) and 0.85 (EO); p<0.001). Consistency across EEG and MEG was excellent (peak frequency: 0.86 (EC); relative power 0.89 (EC) and 0.84 (EO); p<0.001).ConclusionElectrophysiological recordings of brain networks are emerging as a diagnostic tool for early detection of dementia. Good stability of the measured signal features is a prerequisite for a clinically feasible tool. Here, we demonstrate good or excellent stability for prominent spectral features both within and across EEG and MEG recordings.