Data-oriented neuron classification from their parts

Evelyn Perez Cervantes, Cesar Henrique Comin, Roberto Marcondes Cesar Jr., Luciano da Fontoura Costa

Abstract: The shape of a neuron can reveal many interesting properties about its function. Therefore, organizing neuronal cells into appropriate classes according to their respective shape is a fundamental endeavor in neuroscience. Available online datasets allow new data-oriented approaches to solve such neuroscience problems. Here we analyze the feasibility of classifying neurons according not to their respective wholes, but to its constituent parts. Such a study may reveal interesting insights, including whether parts of the neuronal dendritic arborization preserve proper information about the morphology of the whole neuron. Experimental results using open datasets are reported, thus corroborating our approach.