Variability Patterns in Amnestic Mild Cognitive Impairment and Alzheimer’s Disease

Submission No:


Submission Type:

Abstract Submission 


Liwen Zhang1,2,3, Eric Kwun Kei Ng1, Joanna Su Xian Chong1, Hee Youn Shim1, Yng Miin Loke1, Boon Linn Choo1, Narayanaswamy Venketasubramanian4, Boon Yeow Tan5, Christopher Li-Hsian Chen2,3, Juan Zhou1,6


1Duke-National University of Singapore Medical School, Singapore, Singapore, 2Memory Ageing and Cognition Centre, National University Health System, Singapore, Singapore, 3National University of Singapore, Singapore, Singapore, 4Raffles Hospital, Singapore, Singapore, 5St Luke's Hospital, Singapore, Singapore, 6Clinical Imaging Research Centre, the Agency for Science, Technology and Research and National University of Singapore, Singapore, Singapore


Alzheimer's Disease (AD) and its prodromal stage (i.e., amnestic Mild Cognitive Impairment [aMCI]) have been shown to target large-scale functional brain networks. However, functional connectivity derived from fMRI cannot provide information on the temporal variability of blood oxygenation-level-dependent (BOLD) signal amplitude. Optimal levels of moment-to-moment BOLD signal variability (variability hereafter) have been suggested to be important for normative brain functioning and is compromised in ageing brains (Fox and Raichle 2007, Grady and Garrett 2014). It remains largely unknown how variability changes in the course of AD and whether it is associated with disease progression. Therefore, we investigated variability at rest at two characteristic slow frequency bands in AD and aMCI. Given the network disruptions in AD/aMCI, we hypothesized altered variability in AD-related networks in AD and/or aMCI.


We studied 103 participants with aMCI, 98 patients with AD, and 48 age-matched healthy controls (HC). All participants were administered a neuropsychological assessment battery at baseline and 2-year follow-up, assessing two memory and five non-memory domains (Chong, Liu et al. 2017). A global cognition z-score was calculated by averaging all domain-specific z-scores for each participant.

MRI data were acquired from a 3T Siemens Magnetom Tim Trio scanner using a 32-channel head coil, including a whole-brain T1-weighted anatomical image and T2*-weighted resting-state functional images. Preprocessing was performed with a standard pipeline as described previously (Chong, Liu et al. 2017). An index of variability was calculated using in-house matlab scripts based on published methodology (Martino, Magioncalda et al. 2016). Briefly, standard deviation (SD) of the BOLD signal at each voxel was first calculated in the whole band (0-0.25 Hz) and sub-bands (i.e. slow4: 0.027-0.073 Hz; slow5: 0.01-0.027 Hz) separately. Fractional SD (fSD) in each sub-band was then obtained by dividing the sub-band SD by the SD of the whole band. Finally, the fSD maps were spatially normalized across the whole brain, resulting in voxel-wise z-transformed fSD (zfSD) maps at slow4 and slow5 respectively for each participant. Moreover, total grey matter volume (GMV) and hippocampal volume were obtained using voxel-based morphometry (VBM8 toolbox).

To examine whole-brain voxelwise group differences in variability between HC, aMCI and AD, zfSD maps at slow4 and slow5 were subject to separate one-way ANOVAs, with group as the independent variable and age, sex, education years and GMV as covariates. Threshold was set at p<.05 FWE corrected on the cluster level with a voxel-defining threshold of p<.001 (uncorrected).

We next tested the correlation between variability in the regions showing group differences and 1) baseline global cognition, and 2) global cognitive change (baseline - year 2). Given that hippocampal atrophy is a robust AD pathology (van de Pol, Hensel et al. 2006), we also tested whether variability was related to hippocampal volume at baseline. All correlation analyses were restricted to the patient groups (i.e. aMCI and AD combined). Threshold was set at p<.05 (two-tailed).


Compared to AD and HC, aMCI had higher variability in the default mode network (DMN) but lower variability in the salience network (SN) at both bands (Fig. 1&2). Interestingly, higher DMN variability and lower SN variability at both bands were associated with better baseline cognition and greater hippocampal volume across all aMCI and AD patients. More importantly, higher slow 4 variability in the DMN at baseline related to less cognitive decline over a 2-year follow-up across all patients (ps<0.05).
Supporting Image: Figure1.jpg
Supporting Image: Figure2.jpg


The divergent variability changes in the DMN and SN in aMCI participants compared to AD/HC and its relationships with cognition and neurodegeneration suggest a possible compensatory mechanism against further brain and cognitive deterioration at the prodromal stage of AD.

Disorders of the Nervous System:

Alzheimer's Disease and Other Dementias 1

Imaging Methods:


Lifespan Development:


Modeling and Analysis Methods:

Task-Independent and Resting-State Analysis 2


Other - BOLD signal variability; Alzheimer’s disease; Amnestic mild cognitive impairment; Default mode network; Salience network; cognitive decline

1|2Indicates the priority used for review