Automated Processing of Post-Mortem Cortex Images Reveals Physiological Changes Associated with Dementia Sub-types

Automated classification is a well established technique for pattern recognition. In this work, the focus is upon testing for detectable differences between categories, rather than upon building a model for predicting category from new data. We applied image processing and pattern recognition techniques to determine whether there are differentiating features in the structure of blood vessels of the cortex when comparing dementia, its subtypes, and normal cortex. We derived measures from images of tissue using multi fractal analysis, which complements more conventional image analysis techniques. Although statistical methods found no evidence of differences between dementia subtypes, several machine learning methods were able to correctly classify many instances, leading us to conclude that detectable difference do exist. Further investigation along these lines may provide new understanding of the causes, pathology and treatment of these diseases. Our findings demonstrate the utility of multi-fractal analysis combined with machine learning techniques in dementia research.