A Window into Dementia
- Sara Kimmich
- Sep 11, 2017
- 3 min read
1. Description
We'll be analyzing cortical thickness values of normal, pre-clinical and clinical populations in pre-determined fronto-parietal ROIs as a function of age, beta amyloid values and behavioral measures (for the datasets that have made them available).
We'll be using the DLBS dataset as a training set and the ADNI healthy control subsample as a replication set. The HAB dataset will be considered as an explorative pre-clinical set before actually testing the distance of MCI and AD patients from the normal curve.
2.Theoretical Motivation
As most of the literature focuses on memory decline in healthy older individuals and Alzheimer's patients, we aim at extending the current knowledge of cognitive decline to the executive functions' field, by exploring the relation between thickness measures (and brain volume measures) and behavioral performance on executive functions' tasks that are similar across the datasets.
In this way, by performing this analysis on the full spectrum, we will be able to trace a continuous trajectory between the normal, pre-clinical and clinical population as far as cognitive control decline is concerned.
3. Research Design
3.1. Data sets
Training Set: Dallas Lifespan Brain Study (DLBS, n=315, healthy controls HC, age range: 20-89 y/o)
Replication Set: randomly chosen healthy controls from Alzheimer's Disease Neuroimaging Initiative (ADNI, n=150 , HC, age range: 55-90 y/o)
Pre-Clinical Set: Harvard Brain Aging (HBA, n=284, Cognitively Normal CN age range: 60-90 y/o)
Test set: ADNI (n=300, evenly distributed between EMCI,MCI,LMCI and AD, age range: to be defined )
3.2 Pipeline
Quality control using MRIQC tool
Neck removal with FSL robustfov tool
N4 non-uniformity correction using ANTs N4BiasFieldCorrection tool
Run the Freesurfer 6.0 pipelines
Freesurfer Qc → manual assessment with tkmedit for biggest issues: adding control points, white matter filling, editing the pial
Extract cortical thickness measures of fronto-parietal ROIs (to be defined: dLPC, ACC, IPL, IFG?) For cortical ROIs measures we'll be using the Desikan-Killiany atlas and subcortical ROIs (if any): Freesurfer aseg atlas
Extract Ventricular and total brain volume meausures
Fitting of the overlapping curves and computing distances
Research questions:
Are DLBS and ADNI normal controls curves overlapping?
Are the normal curves and the HAB curve any different? Can we use the latter to predict later cognitive impairment?
Can we predict the degree of cognitive impairment(pre-clinical, EMCI, MCI, LMCI, AD) based on the distance of the subjects from the normal curves?
4. Statistical Analysis
4.1. Intra-group analysis
Training Set
Correlation 1: Cortical Thickness ROIs x Age
Correlation 2: Behavioral Performance in the Cambridge Neuropsychological Test Automated Battery x Age
Brain-Behavior correlation: Interaction between age x cortical thickness ROIs x performance in the Cambridge Neuropsychological Test Automated Battery
Replication Set
Correlation 1: Cortical Thickness ROIs x Age
Correlation 2: Behavioral Performance in the Trail Making Test x Age
Brain-Behavior correlation: Interaction age x thickness ROIs x performance TMT
Pre Clinical Set
Correlation 1: Cortical Thickness ROIs x Age
Correlation 2: beta amyloid x Age → high beta amyloid subjects vs low beta amyloid subjects
Beta amyloid values x thickness ROIs x age
Behavioral Performance in Trail Making Test x age
Behavioral Performance in TMT x cortical thickness
Subj with high/low beta amyloid: cortical thickness x Behavioral Perfomance in Trail Making Test x age
Test Set
Extract cortical thickness measures of ROIs
Histogram 1: Cortical Thickness ROIs in each group
Histogram 2: Behavioral Performance in the Trail Making Test x each group
Correlation: Thickness ROIs x Performance in each Group
4.2. Infra-group analysis
We'll be performing t-tests between:
DLBS and ADNI normal controls
DLBS/ADNI normal controls and HAB
DLBS/ADNI normal controls and EMCI
DLBS/ADNI normal controls and MCI
DLBS/ADNI normal controls and LMCI
DLBS/ADNI normal controls and AD
We'll be performing ANOVAs between:
EMCI-MCI-LMCI-AD
normal curve (DLBS/ADNI controls) - pre-clinical curve (HAB) - clinical curve (that includes EMCI,MCI,LMCI and AD)
5. Code Development
Freesurfer recon-all will be run using Python language (nipype). Our statistical analysis will be performed using "rpy2" package in Python.
Contact: dementiawindow on slack
Team: @pog87, @giuliabaracc, @francescap, @ludovica, @vaanathi
Comentarios