Alzheimer disease (a common form of dementia) is one of the heaviest and incurable disease of a human brain, resulting in intellectual loss and low mental activity of a person. The diagnosis of the disease implies a full inspection of the patient including analysis of mental activity and studying of a brain by series of MRI images. Inspecting of MRI images first of all allows to estimate the form of a brain, to reveal infringements in his structure. However on early stages of dementia the morphological changes in a brain structure are not very obvious and often hidden completely.

In the present work a new method for analysis of MRI tomograms of patients with early stages of Alzheimer's disease is proposed. The method is based on soft modeling approach, where different integral and local features of tomogram structure are calculated and used further as predictors in classification models. A degree of illness (0 – patient without dementia, 0.5 – pre-dementia, 1.0 – early dementia), has been used as a class variable.

A priory knowledge of changes in a brain structure during development of dementia allowed us to narrow a range of potential features on a first step of this work. Than a set of numerical experiments have been carried out to the most relevant features. Finally the best results were achieved using statistics of 2D Fourier transform and fractal properties of brain convolutions.