Dynamically Spreading Wave of Gray Matter Loss Visualized in Alzheimer’s Disease using Cortical Pattern Matching and a Brain Atlas Encoding Atrophic Rates

Paul M. Thompson1, Kira Hayashi1, Greig de Zubicaray2, Andrew L. Janke2,3, Stephen E. Rose2, Stephanie Dittmer1, James Semple4, David Herman1, Michael S. Hong1, Michael S. Mega1, David M. Doddrell2, Arthur W. Toga1
1Laboratory of Neuro Imaging, Brain Mapping Division, and UCLA Alzheimer Disease Center, Department of Neurology, UCLA School of Medicine
2Centre for Magnetic Resonance, University of Queensland, Brisbane 4072, Australia
3McConnell Brain Imaging Center, Montreal Neurological Institute, McGill University, Canada
4GlaxoSmithKline Pharmaceuticals plc, Cambridge, UK

We report the mapping of a dynamically spreading wave of gray matter loss in the brains of Alzheimer’s patients, scanned repeatedly with MRI. The loss pattern spread from temporal cortices into frontal and cingulate brain regions as the disease progressed. Deficit patterns were resolved with a novel 4D cortical pattern matching strategy which computes correspondences between cortical gyri across subjects and across time. This procedure separates geometric (shape) and tissue density changes over time, controlling for cortical pattern variation. Systematic features were reinforced in the average maps, and compared with tensor-based measures of atrophic rates encoded in a statistical brain atlas [1],[2].

A population-based brain atlas containing 6840 anatomical surface models was created from 3D MRI (SPGR) scans of 43 AD patients (age: 68.7+/-1.7 yrs.; 24 females/19 males; MMSE score: 20.0+/-0.8) and 34 controls matched for age, education, gender and handedness (all right-handed). After affine alignment of individual data, gyral pattern and shape variations were encoded using high-dimensional elastic deformation mappings [3] driving each subject’s cortical anatomy into a group average configuration. Dynamic maps of atrophic rates, with millions of degrees of freedom, were generated for 17 AD patients and 14 matched controls scanned longitudinally (interscan interval: 2.6±0.3 yrs.; final age: 71.3±1.8 yrs.). To create maps of change, parametric surface models of cortical, hippocampal, ventricular, and callosal systems drove an elastic warping field reconfiguring the earlier scan’s anatomy onto the later one. Local volume loss was quantified. Changes in cortical gray matter density were mapped by computing warping fields that matched cortical patterns across hemispheres, across time, and across subjects. Annualized 4D maps of tissue loss rates within each subject [4] were elastically realigned for averaging across diagnostic groups, and covaried for age, sex, hemisphere, and MMSE score. Statistics of local loss rates were visualized using color-coded maps; significance was assessed using a covariant partial differential equation [5] to induce a grid on the population average cortex whose deformation gradient matched the smoothness tensor of the residuals. Permutation was applied to features in its parameter space.

A dynamically spreading wave of cortical gray matter loss (up to 4-5%/year locally) was detected in the AD patients, beginning in temporal cortices. Primary sensorimotor cortices were comparatively spared as the loss pattern shifted into frontal cortices, and intensified in the cingulate and paralimbic belts. Corpus callosum curvature increases were significant in controls (p<0.05), higher in AD (p<0.05), and were strongly correlated with hippocampal loss rates in both groups (pooled p<0.002 left, p<0.000004 right; r=0.54,0.73). In AD, greatest dynamic change rates were found in the inferior ventricular horns (L:+14.7+/-5.8%/yr.; R:+16.3+/-3.5%/yr.), with significant expansion rates bilaterally even in controls (L:+3.7+/-1.2%/yr.; R:+1.7+/-1.2%; p<0.001,p<0.01).

These dynamic maps show promise in charting the dynamic progress of Alzheimer’s disease, and reveal a changing pattern of deficits. We will use them in future to detect where deficit patterns are modified by drug treatment and known risk genotypes.

[1]. Janke et al. Proc. ISMRM 580(2000); [2]-[5]: Thompson et al., Nature 404(6774):190-193(2000); Nature Neuroscience 4(12):1253-8(2001); PNAS 98(20):11650-11655(2001); HBM 9(2):81-92(2000).