Sixty-three preclinical AD subjects were included in this study. They were recruited from the normal control volunteers of the Catholic Dementia Brain Imaging Database, which holds brain scans of outpatients and inpatients at the Department of Geriatric Psychiatry, the Saint Vincent’s Hospital, the Catholic University of Korea from 2010 to 2016. The inclusion criteria of the Exercise group were as follows: 1) subjects aged ≥60 years; 2) Mini-Mental Status Examination score ≥27; 3) Clinical Dementia Rating=0 [18]; 4) amyloid retention corresponds to global mean standard value uptake ratio (SUVR) ≥1.4. 5) Doing moderate intensity aerobic exercise (e.g., mountain climbing, dancing, and swimming) more than one hour per day, 5 days per weeks [19]. Most of the inclusion criteria of the Non-Exercise group were the same as the Exercise group [1)-4)], except for the criteria that the Non-Exercise group had a history of doing walking less than 10 min or no exercise per day, 5 days per weeks were included [19]. The cognitive functions of the subjects were evaluated with the Korean version of Consortium to Establish a Registry for Alzheimer’s Disease (CERAD-K) [20], which assess the following cognitive domains: verbal fluency, 15-item Boston naming test (BNT), the Korean version of Mini Mental Status Examination (MMSE-K) [21], constructional praxis (CP), word list memory (WLM), word list recall (WLR), word list recognition (WLRc), constructional recall (CR). The study was conducted in accordance with the ethical and safety guidelines set forth by the local Institutional Review Board of the Catholic University of Korea and written informed consent was obtained from all study subjects.

FBB [(F-18) florbetaben] was produced and FBB-PET data were collected and analyzed as previously described [22]. Each individual participant’s MRI was utilized for co-registration and defining the ROI and for correcting partial volume effects from expanding cerebrospinal spaces accompanying cerebral atrophy [23,24]. Analysis of the FBB PET data utilized a standardized uptake value ratio (SUVR) 90 min post-injection, using the cerebellar cortex region of interest as the reference. Global Aβ burden was expressed as the average SUVR of the mean for the following 5 cortical ROIs: frontal, superior parietal, lateral temporal, and anterior and posterior cingulate cortex/precuneus as described in previous study [24].

The FBB PET data were acquired within 4 weeks of clinical screening and cognitive function test. We used a cut-off for ‘high’ or ‘low’ neocortical SUVR of 1.4, consistent with cut-off values used in previous FBB-PET study [25].

Imaging data were collected at the Department of Radiology, St Vincent’s Hospital, The Catholic University of Korea, using a 3T Siemens Verio machine and eight channel Siemens head coil (Siemens Medical Solutions, Erlangen, Germany). The parameters used for the T1-weighted volumetric magnetization-prepared rapid gradient echo scan sequences were TE=2.5 ms, TR=1,900 ms, inversion time=900 ms, FOV=250 mm, matrix=256×256, and voxel size=1.0×1.0×1.0 mm [3].

For cortical reconstruction and volumetric segmentation of the whole brain, Freesurfer image analysis suite (version 6.0, http://surfer.nmr.mgh.harvard.edu), which is documented and freely available online, was used. The technical details of these procedures have been described in previous publications [26,27]. Briefly, the processing stream includes a Talairach transform of each subject’s native brain, removal of the non-brain tissue, and segmentation of the gray matter/white matter (GM/WM) tissue. The cortical surface of each hemisphere was inflated to an average spherical surface to locate both the pial surface and the GM/WM boundary. The entire cortex of each subject was visually inspected, and any topological defects were corrected manually, blind to the subject’s identity. The cortical thickness was computed as the shortest distance between the pial surface and the GM/WM boundary at each point across the cortical mantle. The global mean cortical thickness for each subject was computed by averaging the cortical thickness at each vertex, right and left hemispheres separately, and was used in the statistical analyses. The regional thickness value at each vertex for each subject was mapped to the surface of an average brain template allowing visualization of data across the entire cortical surface (described at http://surfer. nmr.mgh.harvard. edu/fswiki/FsAverage). In addition, the entire cerebral cortex was parcellated into 34 regions [28,29], and a variety of surface-based data, including maps of cortical volume and surface area as well as curvature and sulcal depth, were created. Data were resampled for all subjects onto a common spherical coordinate system [30]. The cortical map of each subject was smoothed with a Gaussian kernel of 10-mm full width at half-maximum for the entire cortex analyses. The subcortical volumes were obtained from the automated procedure for volumetric measures of the brain structures implemented in Freesurfer. In all, 27 volumetric measures were investigated and extracted seven subcortical structures (white matter, caudate, thalamus, pallidum, putamen, hippocampus, and amygdala) from each hemisphere.

Statistical analyses for demographic data were performed with the Statistical Package for Social Sciences software (SPSS, version 20.0, IBM Corp., Armonk, NY, USA). Assumptions for normality were tested for all continuous variables. Normality was tested using the Kolmogorov-Smirnov test. All variables were normally distributed. The independent t-test and the χ² test were used to assess potential differences between the Exercise groups and Non exercise groups for all demographic variables. All statistical analyses had a two-tailed a level of <0.05 for defining statistical significance. The general linear model (GLM) was implemented at each vertex in the whole brain to identify the brain regions in which the Exercise groups showed significant differences in cortical thickness relative to Non exercise group, using the FreeSurfer’s mri_glmfit (described at http://surfer.nmr.mgh.harvard.edu/fswiki/mri_glmfit). The All analyses were performed for the right and left hemispheres separately. The threshold was set at p<0.05 [false discovery rate (FDR)] to resolve the problem of multiple comparisons. The seven subcortical structure volumes (i.e., total white matter volumes, thalamus, caudate nucleus, putamen, pallidum, hippocampus, and amygdala) were imported into the SPSS 20.0 software for statistical analyses (IBM Corp.). To assess the main effects of diagnosis on the volume of subcortical structures, we used analysis of covariance (ANCOVA) with TIV, education, gender, and age as nuisance variables.