Brain Images of Normal Subjects (BRAINS) Bank David Alexander Dickie Dr Dominic E. Job
Background Age and disease affect brain structure The effects are disparate Much MRI data are needed ~300 normal ageing (>60yrs) subjects “Atlases” of the aged brain are limited
Background Brain Images of Normal Subjects (BRAINS) bank >1000 normal sbjs >60yrs BRAINS models and atlases calculate distributions (not assume) Data requires much image processing Pilot ~200 normal, ~200 AD sjbs, 60-94 years
MR Image processing
Brain extraction
Brain extraction Brain Extraction Tool (BET) commonly used
Brain extraction Brain Extraction Tool (BET) commonly used
Template based brain extraction Advanced Normalization Tools (ANTS) http://www.picsl.upenn.edu/ANTS/ Uses diffeomorphic (super nonlinear) spatial normalisation
Image registration
ANTS diffeomorphic spatial normalisation
ANTS diffeomorphic spatial normalisation
ANTS diffeomorphic spatial normalisation But catastrophes still happen
ANTS diffeomorphic spatial normalisation ANTS takes ~1 hour per subject (computer) Still requires by slice checking ~10 minutes checking per subject 460 subjects took ~2.5 months Catastrophes still happen >1000 subjects in full-scale study
Data driven brain volume models Statistical models oft used in brain imaging The general linear model (GLM) Assume data generation and distribution Transformations lose natural data, have risks, complexity Image banks support data driven models
Brains are heteroscedastic
See, they’re different
Data driven vs. general models DDPM has ~65% less error
Data driven brain voxel models Statistical voxel based morphometry (VBM) The general linear model (GLM) Assumes data generation and distribution Transformations lose natural data, smoothing Image banks support data driven models
BRAINS atlases BRAINS MNI 152
Image registration
BRAINS atlases 1
BRAINS atlases Normal Alzheimer’s Red=95th
White matter lesions
White matter lesions Percentile 95 75 50 25 5 A
White matter lesions Percentile 95 75 50 25 5 A
Percentiles of grey matter density in a normal ageing and Alzheimer’s disease subject 5th 25th 50th 75th 95th Alzheimer’s disease has lowest percentiles of GM in MTL
Percentiles of grey matter density in a normal ageing and Alzheimer’s disease subject Bad OK Good Alzheimer’s disease has lowest percentiles of GM in MTL
Median percentile image of grey matter density in Alzheimer’s disease (n=49) and control (n=49) subjects Alzheimer’s Control Percentile 5th 25th 50th 75th 95th Alzheimer’s has lowest percentiles of GM across the cortex, specifically hippocampus.
BRAINS vs. VBM
BRAINS vs. VBM Needs data Less assumptions Anatomical resolution Specific anatomy Individuals Size of differences Simpler
SINAPSE SPIRIT, MRC, Tony Watson Thank you