A Window into Dementia

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

1. Training Set: Dallas Lifespan Brain Study (DLBS, n=315, healthy controls HC, age range: 20-89 y/o)

2. Replication Set: randomly chosen healthy controls from Alzheimer's Disease Neuroimaging Initiative (ADNI, n=150 , HC, age range: 55-90 y/o)

3. Pre-Clinical Set: Harvard Brain Aging (HBA, n=284, Cognitively Normal CN age range: 60-90 y/o)

4. Test set: ADNI (n=300, evenly distributed between EMCI,MCI,LMCI and AD, age range: to be defined )

3.2 Pipeline

1. Quality control using MRIQC tool

2. Neck removal with FSL robustfov tool

3. N4 non-uniformity correction using ANTs N4BiasFieldCorrection tool

4. Run the Freesurfer 6.0 pipelines

5. Freesurfer Qc → manual assessment with tkmedit for biggest issues: adding control points, white matter filling, editing the pial

6. 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

7. Extract Ventricular and total brain volume meausures

8. Fitting of the overlapping curves and computing distances

Research questions:

1. Are DLBS and ADNI normal controls curves overlapping?

2. Are the normal curves and the HAB curve any different? Can we use the latter to predict later cognitive impairment?

3. 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

1. Correlation 1: Cortical Thickness ROIs x Age

2. Correlation 2: Behavioral Performance in the Cambridge Neuropsychological Test Automated Battery x Age

3. Brain-Behavior correlation: Interaction between age x cortical thickness ROIs x performance in the Cambridge Neuropsychological Test Automated Battery

• Replication Set

1. Correlation 1: Cortical Thickness ROIs x Age

2. Correlation 2: Behavioral Performance in the Trail Making Test x Age

3. Brain-Behavior correlation: Interaction age x thickness ROIs x performance TMT

• Pre Clinical Set

1. Correlation 1: Cortical Thickness ROIs x Age

2. Correlation 2: beta amyloid x Age → high beta amyloid subjects vs low beta amyloid subjects

3. Beta amyloid values x thickness ROIs x age

4. Behavioral Performance in Trail Making Test x age

5. Behavioral Performance in TMT x cortical thickness

6. Subj with high/low beta amyloid: cortical thickness x Behavioral Perfomance in Trail Making Test x age

• Test Set

1. Extract cortical thickness measures of ROIs

2. Histogram 1: Cortical Thickness ROIs in each group

3. Histogram 2: Behavioral Performance in the Trail Making Test x each group

4. Correlation: Thickness ROIs x Performance in each Group

4.2. Infra-group analysis

We'll be performing t-tests between:

1. DLBS and ADNI normal controls

2. DLBS/ADNI normal controls and HAB

3. DLBS/ADNI normal controls and EMCI

4. DLBS/ADNI normal controls and MCI

5. DLBS/ADNI normal controls and LMCI

6. DLBS/ADNI normal controls and AD

We'll be performing ANOVAs between:

1. EMCI-MCI-LMCI-AD

2. 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

View Project Github Repository