Recommendations for straightforward and rigorous methods of counting neurons based on a computer simulation approach.

Any investigation of the total number of neurons in a given brain region must first address the following questions. What is the best method for estimating the total number of neurons? What are the validity and the expected precision of the obtained data? What precision must the estimates attain with respect to the scientific question? In the present study, these questions were addressed using a computer simulation. Virtual brain regions with various spatial distributions of virtual neurons were modeled. The total numbers of virtual neurons in the modeled brain regions were repeatedly estimated by simulation of modern design-based stereology, either by using the 'fractionator' method or by the established method based on the product of estimated neuron density and estimated volume of the reference space. We show that estimates of total numbers of neurons obtained using the fractionator are from a statistical and economical standpoint more efficient than corresponding estimates obtained using the density/volume procedure. Furthermore, the use of two simple prediction methods (one for homogeneous and the other for clustered neuron distributions) permits satisfactory predictions about the variation of presumably any estimates of total numbers of neurons obtained using the fractionator. Finally, we show that assessing the reliability of estimates of mean total neuronal numbers using the ratio between the mean of the squared coefficients of error of the estimates and the squared coefficient of variation of the estimated total neuronal numbers, a frequently employed method in stereological studies, is neither useful nor informative. The present results may constitute a new set of recommendations for the rigorous usage of design-based stereology. In particular, we strongly recommend counting considerably more neurons than is currently done in the literature when estimating total neuronal numbers using design-based stereology.