Laboratory of Neuro-Imaging and Brain Mapping Division, Dept.
Neurology, UCLA School of Medicine
Alzheimer's Disease Center, Dept. Neurology, UCLA School of Medicine
Summary. We report the first, detailed, population-based maps of early gray matter loss across the cortex in Alzheimer's Disease (AD), identified with a new computational strategy that uncovers disease-specific patterns of cortical organization and tissue distribution. The approach, based on random tensor fields, represents information on the variability in cortical patterns and individual gray matter distribution in a probabilistic brain atlas specialized for the disease. With a new method for averaging cortical anatomy, disease-specific features and cortical asymmetries emerged that were not observed in individual subjects due to normal anatomic complexity. Fundamental patterns in the structural variability of the human cortex were resolved, revealing unsuspected directional biases in cortical pattern variation in both healthy and diseased populations. Population-based maps revealed earliest gray matter loss in temporo-parietal cortices, with left greater than right hemisphere deficits, and a comparative sparing of sensorimotor and occipital cortices.
Methods. High-resolution 3D SPGR (spoiled GRASS) MRI volumes were acquired from 26 subjects with mild Alzheimer's Disease (AD; age: 75.8±1.7 yrs.; MMSE score: 20.0±0.9), and 20 normal elderly controls (72.4±1.3 yrs.) matched for age, sex, handedness and educational level. 84 anatomical models per brain were created for all 46 subjects (16 deep sulcal, callosal and hippocampal surfaces, all major cortical sulci, Sylvian fissures, 14 ventricular regions, and 36 gyral and functional boundaries). A nearest-neighbor tissue classifier was also used to create maps of gray matter, white matter and CSF. After affine alignment of the individual data to a group average size and shape , high-dimensional elastic mappings  were computed to reconfigure the individual MRI data into a group mean configuration . The reconfigured individual data were then averaged on a voxel-by-voxel basis to create a group-specific average brain image template with well-resolved features at the cortex. Because cortical patterns were exactly matched across subjects, the disease-specific average anatomical representation exhibited highly-resolved structures in their mean spatial locations. Statistical variations in cortical patterning, asymmetry, gray matter distribution and average gray matter loss were encoded as a non-stationary Gaussian random tensor field, and used to detect disease-specific differences. Elastic matching was also used to associate gray matter measures from homologous cortical regions. Because the theory of stationary Gaussian fields may not be directly applicable for assessing group differences in structural imaging data , an approach related to statistical flattening  was developed to assess the statistical significance of gray matter loss across the cortical sheet in Alzheimer's Disease. First, a partial differential equation:
gij(d2u/dridrj) + d/duj(Sij)uri = 0
was run in the parameter space of the group average cortex to generate a deformed grid u(r) whose deformation gradient tensor matched the smoothness tensor Sij of the residuals of the statistical model (here gij is the contravariant metric tensor of the grid). Relative to this new computational grid, the residuals became stationary and isotropic, and p-values for the gray matter reductions were evaluated.
Results. In the population-based maps, an overall pattern of left greater than right hemisphere gray matter loss was observed, with sharp contrasts between heteromodal and idiotypic cortex. After affine normalization, the principal directions of both normal and diseased gyral pattern variability exhibited clear boundaries, demarcating cortical zones appearing at different embryonic phases (cf. ). In controls, peak 3D r.m.s. structural variability values in left perisylvian (14 mm) and right frontal association cortex (8-14 mm) contrasted with the highly stable cortical regions between the central and postcentral sulci (2-8 mm), collateral and occipito-temporal sulci (2-9 mm), the interhemispheric margin (2-10 mm), olfactory sulci (2-4 mm) and temporal poles (2-4 mm). Higher anatomical variability values in AD may reflect additional disease-related atrophic change. Once these variations were controlled, pervasive early gray matter loss (up to 30% locally) in temporo-parietal association cortices was visualized in the Alzheimer's disease group (p<0.05). This early gray matter reduction was found to be considerably greater than in sensorimotor and occipital cortices (0%-5% locally, p > 0.05), consistent with cognitive, metabolic and histologic patterns in early AD.
Conclusion. The potential relevance of this probabilistic atlasing approach for uncovering disease-specific features of anatomy, for mapping cerebral asymmetry, and for understanding the structural and functional variability of the human cortex, cannot be overlooked. In addition, the accurate mapping of gray matter changes in a living population with Alzheimer's Disease holds significant promise for genetic, longitudinal and interventional studies of dementia.
References. : Woods RP et al., Human Brain Mapping 8(2-3):73-9(1999). [2,3]: Thompson PM et al., IEEE-TMI 15(4):402-417(1996); Human Brain Mapping 9(2), Jan. 2000 [in press]. : Worsley KJ et al., Human Brain Mapping 8(2-3):98-101(1999).
|RESUME| E-MAIL ME| PERSONAL HOMEPAGE| PROJECTS|