Automatic classification of retinal ganglion cells

Abstract

Experiments on the classification of neural cells by means of 2D shape analysis and pattern recognition procedures are described. Two types of the cat's retinal ganglion cells (/spl alpha/ and /spl beta/ cells) are characterized by a set of morphological and specific features, which are analyzed by feature-ordering techniques. The shape features include the fractal dimension, the normalized multiscale bending energy as well as standard measures such as the size of the soma and of the dendritic arborization. Several classification experiments using two statistical classifiers (k-nearest neighbor and maximum likelihood) were carried out based on the information provided by the a priori feature-ordering tests. Encouraging recognition results are reported.