This retrospective study used the qEEG and three-dimensional T1-weighted MRI (3D T1 MRI) data of patients who visited the Chung-Ang University Hospital Department of Neurology from January 2012 to May 2019 and were diagnosed with single-domain amnestic MCI. This study was approved by the institutional review board of our center (IRB number 1802-004-16143). Written informed consent was obtained from all participants.
Participants were aged 55 years or older, underwent 3D T1 MRI and qEEG within 2 weeks, and met the single-domain amnestic MCI criteria. The criteria were as follows: (1) presence of memory complaints; (2) intact performance of activities of daily living; (3) objective verbal memory impairments on the Seoul Neuropsychological Screening Battery (at least 1.0 SD below age- and education-adjusted norms); (4) Clinical Dementia Rating of 0.5 (1); and (5) not demented according to the Diagnostic and Statistical Manual of Mental Disorders (DSM)-IV criteria. Subjects were divided into an EF MCI and an RF MCI group. EF was defined as both delayed recall and recognition scores on a verbal learning test below 1.0 SD; RF was defined as only a delayed recall score below 1.0 SD. The subjects in EF MCI group may be accompanied by retrieval failure. Resting-state EEG data were obtained from all 165 patients with amnestic MCI comprising 87 with EF and 78 with RF. 3D T1 MRI data were available for brain volume analysis for 147 of all subjects with amnestic MCI (for 78 with EF and 69 with RF) because of problems in the preprocessing of images (Figure 1).
The 3D T1 MRI imaging data of 71 age-matched cognitive normal control (NC) subjects were selected from the repository. The inclusion criteria for NCs were as follows: (1) from a community-based population; (2) no abnormalities based on a health screening questionnaire ; (3) absence of memory complaints; (4) a Korean Dementia Screening Questionnaire score ≤ 6 ; (5) a Mini-Mental State Examination (MMSE) score > 26; (6) intact activities of daily living (K-IADL ≤ 0.42); (7) no history of thyroid dysfunction, vitamin B12 deficiency, or folate deficiency; and (8) at least 6 years of education.
No participant presented any structural abnormalities on MRI, such as territorial infarctions, intracranial hemorrhage, brain tumors, or hydrocephalus; lacunar infarcts or mild to moderate subcortical or periventricular white matter hyperintensities did not lead to exclusion. Patients with major psychiatric disease, such as schizophrenia, major mood disorder, and chronic alcoholics were also excluded.
2.2 qEEG analysis
Resting-state EEG was conducted using the standard 10–20 system (21 electrodes) and a digital electroencephalograph (Comet AS40 amplifier EEG GRASS; Telefactor, USA) (Jaspers, 1958), and all electrodes were referred to linked ear references. Electrode skin impedance was always below 5 kΩ. The EEG signal was analog-filtered with a band pass of 0.5–70 Hz and digitized and stored on magnetic disks for further analysis. EEG sampling was conducted with eyes open for 30 seconds and with eyes closed for 30 seconds, 10 times, at a rate of 200 Hz. Of these, about 3 minutes of eyes-closed data was used. One epoch is 4 seconds long, and an average of 45 epochs were analyzed. The measured eyes-open and eyes-closed data were converted according to the linked ear reference and stored in text format without filtering. While resting-state EEG data were recorded, patients were lying down in a resting position in a sound-attenuated room. EEG noise preprocessing and group analyses were conducted using iSyncBrain™ (iMediSync, Inc., Korea) (https://isyncbrain.com/), a cloud-based, artificial intelligence EEG analysis platform. The eyes-closed EEG segments were uploaded to iSyncBrain™. Prior to data analysis, artifacts in the raw data were removed by visual inspection and an advanced mixture independent component analysis (amICA) . qEEG features were obtained at the sensor and at the source level. At the sensor level, relative power at eight frequency bands (delta [1–4 Hz], theta [4–8 Hz], alpha1 [8–10 Hz], alpha2 [10–12 Hz], beta1 [12–15 Hz], beta2 [15–20 Hz], beta3 [20–30 Hz], and gamma [30–45 Hz]) was calculated using a power spectrum analysis. In the source level analysis, the current distribution across the brain was assessed using the standardized low resolution brain electromagnetic tomography technique , to compare relative power values in 68 regions of interests (ROIs)  and connectivity (the imaginary part of coherency)  between ROIs. EEG coherence has been studied as a measure of brain connectivity , and the imaginary part of coherency (iCoh) has been introduced to avoid volume conduction artifacts . We calculated the connectivity of each of the regional pairwise of 68 ROIs with remaining all other 67 ROIs. We have estimated the functional connectivity at eight frequency bands (delta [1–4 Hz], theta [4–8 Hz], alpha1 [8–10 Hz], alpha2 [10–12 Hz], beta1 [12–15 Hz], beta2 [15–20 Hz], beta3 [20–30 Hz], and gamma [30–45 Hz].
2.3 MRI Volumetry
To determine gray matter (GM) volume changes underlying EF and RF in amnestic MCI, we conducted voxel-based morphometry (VBM) on MRI scans acquired on 3-T scanners manufactured by Philips (Achieva, Amsterdam, the Netherlands). The data were analyzed using the Computational Anatomy Toolbox (CAT12) running on Statistical Parametric Mapping software (SPM12). CAT12 is a VBM toolbox designed by The Structural Brain Mapping Group at the University of Jena (Jena, Germany). First, the DICOM files were converted into nifti format, using MRICRON software (http://people.cas.sc.edu/rorden/mricron/index.html). VBM preprocessing was performed using the default settings of the CAT12 toolbox and the “East Asian Brains” ICBM template. Imaging files were normalized using an affine model, followed by non-linear registration, corrected for bias field inhomogeneities, and then segmented into GM, white matter (WM), and cerebrospinal fluid (CSF) components. The segmented scans were normalized into standard Montreal Neurological Institute space using the Diffeomorphic Anatomic Registration Through Exponentiated Lie (DARTEL) algebra algorithm. The modulation process on the normalized, segmented images consisted of a non-linear deformation, which corrects individual differences in brain size. We reviewed morphological abnormalities and applied smoothing processes to all segmented, modulated, and normalized GM images using an 8-mm full-width-half-maximum Gaussian filter.
2.4 Statistical analysis
To compare demographic and cognitive assessment results between groups, Student’s t-tests for continuous variables were performed. We used IBM SPSS version 25 (IBM, Armonk, NY, USA) for all analyses. Statistical significance was set at p < 0.05. The obtained qEEG features were analyzed according to statistically significant differences between the EF and RF groups (p < 0.05): the power spectra for each of the 19 channels; the source power of 68 ROIs and their connectivity. The all processing including denoising using an amICA, sensor and source level qEEG feature extraction were performed on iSyncBrain ™.
To demonstrate GM volume changes underlying EF and RF in amnestic MCI, we conducted a comparison with processed MR images of cognitively normal subjects using Student’s t-tests. Age and total intracranial volume (TIV), that is, the sum of the GM, WM, and CSF volumes, were classified as nuisance covariates in the GM volume comparisons between the groups. We used a VBM analysis to demonstrate significant atrophic GM areas in the two types of patients with amnestic MCI. To detect GM volume differences between patients with EF and those with RF, t-tests of the VBM on SPM package were also conducted on the processed images. Age and TIV were again classified as nuisance covariates in the GM volume comparisons between groups. Absolute threshold masking was used at a threshold of 0.1. Results were considered statistically significant at p < 0.05 and were corrected for family-wise errors to avoid multiple-comparison problems.