Introduction: Differences in brain structure on magnetic resonance imaging (MRI) between normal ageing and early dementia are subtle. Without serial imaging, these differences may be highlighted by comparing individual scans to templates of normal brain structure. Previous templates had limitations, e.g. based on parametric assumptions that are potentially not met in brain structure data. Therefore, we describe the development of a nonparametric voxel-based method for determining the boundaries of normality, and how this might inform future templates.
Methods: We used structural brain MRI from 236 normal subjects and 138 subjects diagnosed with Alzheimer's Disease (AD) (55-90 years) from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and the Open Access Series of Imaging Studies (OASIS). Half (49) of the normal subjects from OASIS and 49 normal subjects from ADNI were used to create nonparametric voxel-based templates of grey matter (GM), white matter (WM), and cerebrospinal fluid (CSF), according to the calculated values of the 2.5th, 25th, 50th, 75th, and 97.5th percentile ranks of GM, WM or CSF in each voxel. For comparison, we also created a parametric voxel-based template of GM. In a validation study, we ranked each voxel from the remaining normal and AD subjects with the nonparametric templates, and compared these to the parametric methods.
Results: The nonparametric GM template identified areas of atrophy known to be abnormal in AD, e.g. the hippocampus. These are shown on colour-coded brain images. There were differences between the nonparametric and parametric methods.
Conclusion: Nonparametric voxel-based brain ranking may highlight subtle brain structure abnormalities and allows comparison with age-appropriate templates. This may increase the utility of brain imaging in supporting diagnoses of neurodegenerative disorders. The development of robust templates of normal brain structure using nonparametric methods will require large numbers of normal subjects (see: http://http://www.sinapse.ac.uk/research-resources/brains-project).