Abstract
Purpose:
The rat optic nerve axons come from the retinal ganglion cells. Only a few of them have their origin in the pretectal area. The diameters of the optic nerve axons are far from being a homogeneous population. There are at least three main groups of ganglion cells in rat retinas and three different axon types in the rat optic nerve. The morphometric study of optic nerve fibers is a useful tool to research the function, development, aging and pathological conditions of them. Thus, classification of the optic nerve axons is crucial in order to make experimental comparisons. The aim of this work is to classify the optic nerve axons by analyzing their ultrastructural parameters with Artificial Intelligence (AI) methods.
Methods:
Adult albino Wistar rats were anesthetized, transcardially perfused and its optic nerves removed and processed for ultrastructural microscopy studies. Optic nerve axons were analyzed with a computer-linked planimeter. Several parameter, were obtained for each axonal cross-section using a computer program. The parameters were: axon diameter, axon area, myelin sheath thickness, G-Ratio, microtubule number (MTn), neurofilament number (NFn) and R-proportion (R = NFn/[NFn+MTn]). Data were processed with a set of AI methods, two supervised techniques, Multilayer Perceptron (MLP) and Decision Trees (DT), and other unsupervised one, K-Means clustering. All the computations of the decision tree were developed using the WEKA software (Machine Learning Group at the University of Waikato).
Results:
Using k-means clustering analysis we were able to identify three different groups of fibers in the optic nerve, which are consistent with the results obtained in functional studies. The main parameters that allowed us the classification of optical nerve fibers were axon diameter, G-Ratio, and R-proportion. Using these parameters, we analyzed the classification accuracy of MLP and DT, as tools to develop an automated system. In both cases, the Classification Accuracy was above 95%, being higher with the MLP technique that reached a 98.9% of accuracy.
Conclusions:
Results show that morphometric parameters can be used to identify different populations of nerve fibers, with a high accuracy when AI methods are used. Only a limited number of parameters are needed to produce a consistent classification. The MLP technique is the most useful.