Comparing histological data from different brains: sources of error and strategies for minimizing them.

摘要

The recent development of brain atlases with computer graphics templates, and of huge databases of neurohistochemical data on the internet, has forced a systematic re-examination of errors associated with comparing histological features between adjacent sections of the same brain, between brains treated in the same way, and between brains from groups treated in different ways. The long-term goal is to compare as accurately as possible a broad array of data from experimental brains within the framework of reference atlases. Main sources of error, each of which ideally should be measured and minimized, include intrinsic biological variation, linear and nonlinear distortion of histological sections, plane of section differences between each brain, section alignment problems, and sampling errors. These variables are discussed, along with approaches to error estimation and minimization in terms of a specific example-the distribution of neuroendocrine neurons in the rat paraventricular nucleus. Based on the strategy developed here, the main conclusion is that the best long-term solution is a high-resolution 3D computer graphics model of the brain that can be sliced in any plane and used as the framework for quantitative neuroanatomy, databases, knowledge management systems, and structure-function modeling. However, any approach to the automatic annotation of neuroanatomical data-relating its spatial distribution to a reference atlas-should deal systematically with these sources of error, which reduce localization reliability.