Recognition and classification of structure by means of stereological methods in neurobiology

The problem of measuring fine structure of individual particles without losing information on largescale characteristics of particle arrangement is discussed. With the help of stereological and pattern recognition methods a possible solution of this important problem is introduced. The domain specimen for which the brain serves as an example is characterized by two main properties. One of them being the position dependent 'aggregate characteristics' (distribution of neurons within the specimen), the other position invariant 'single cell characteristics' (structural properties of single neurons). It is shown that by simultaneous observation both properties together cannot be detected with sufficient accuracy by conventional methods. This is the decisive problem of 'correlation microscopy'. The method described in this paper is based on selection of the most informative variables and selection of subdomains (e.g. sections and reference planes). This results in a very general probabilistic concept in modern stereology, offering solutions to complex structure classification problems in biology.