Textural 3-dimensional multiscale analysis of MRI volumes of the brain

We describe a method for textural feature extraction of MRI volumes of the brain and, based upon those features, a method for classification and assessment of the anatomical malformations of the brain, due to Alzheimer's Disease (AD). In our research, we make the hypothesis that there is enough detectable textural evidence from a 3D analysis of MR images of the brain to detect and identify the earliest structural changes of AD. To uniquely characterise structural malformations we construct a database of statistical information for 3D textures at different scales, using wavelet operators. The major goal at this stage of our research is to explore the inherent constraints imposed by the structure of the texture and its symbolic description. Our representation benefits from a unique method of parameter reduction, which gives an unambiguous description of the textures of the brain in 3D. One of the key attributes of this model is that, in the case of conflicting statements, it generates a low confidence estimate, thus allowing a local measure of reliability.