Morphological Properties of the Two Types of Caudate Interneurons: Kohonen Self-Organizing Maps and Correlation-Comparison Analysis

摘要

Our previous study found that caudate and putaminal interneurons are morphologically very different, and that accordingly they could be divided in two separate clusters. In addition, it also demonstrated, as a collateral result, that the caudate cluster itself consists of two clusters of morphologically different interneurons. Hence, the objective of this study is a morphological description and subtle differing of morphologies of these two types of caudate interneurons, i.e., an investigation of those morphological traits which characterize them uniquely, and which would distinguish them. Binary two-dimensional images of caudate interneurons, taken from deceased adult human subjects, were analyzed by using 46 parameters, describing the morphology of interneurons. The parameters can be divided in the following classes: size (surface) of a neuron, neuronal shape, length of neuronal morphological compartments, dendritic branching, morphological organization, and complexity. The morphological determination of caudate interneurons was performed in a step-wise manner. The first step was the assignment of each individual neuron to an adequate cluster where it belonged according to morphological criteria. This was done by using the trained artificial neural network, Kohonen self-organizing map. After the clusters were formed, the analysis is further continued by the precise, feature-wise determination of morphological differences found between clusters of caudate interneurons and then finished by defining correlation-based, mutual, inter-parametric relations for each of the clusters. The first was performed by using single-factor analysis, and the second by correlation-comparison analysis. Single-factor analysis showed significance for 34 parameters (morphological features) that distinguish between the clusters. Correlation-comparison analysis extended the results of single-factor analysis by demonstrating significance for 198 inter-parametric correlation pairs that repres