Exploiting distance metrics-based similarity for spatial feature analysis: Application to brain magnetic resonance imaging

The well-known degenerating neurological diseases that cause dementia include Alzheimer's disease (AD), Huntington's disease (HD) and Pick's disease (PD). The spatial features of whole-brain neuroimages depicted by AD, HD and PD diseases have been little explored and thus allow new directions in research. In this study we explore the possibility of distinguishing between patients with neurological disorders and the healthy or normal cognitive (NC) elderly, paying special attention to statistical similarity measurements through histogram analysis. The whole-brain spatial histogram and 2D-texture-descriptor local binary pattern based on rotation invariance are utilised. The histogram comparison by means of the probability of different grey-level appearance needs less computational requirements and has satisfactory recognition accuracy. Various pseudo-metrics for comparing histogram distributions, as well as the data from 21 NC, 24 AD, 18 HD and 16 PD brain magnetic resonance images, were used. The statistical analysis, which is associated with correlation and concordance correlation coefficients, compares pairs of diseases in order to assess the effectiveness of texture features in differentiating between patients with dementia and the healthy elderly.