Neuromorphic Networks as Revealed by Features Similarity

TL;DR

This paper introduces a novel network-based approach using coincidence similarity to classify neuronal cell types based on morphological features, enhancing pattern recognition and revealing interrelationships among cells.

Contribution

It presents a new method employing coincidence similarity for mapping neuronal morphology data into networks, improving classification and understanding of neuronal diversity.

Findings

Well-separated neuronal groups identified

Networks illustrate merging of cells and groups

Enhanced pattern recognition performance

Abstract

The study of neuronal morphology is important not only for its potential relationship with neuronal dynamics, but also as a means to classify diverse types of cells and compare than among species, organs, and conditions. In the present work, we approach this interesting problem by using the concept of coincidence similarity, as well as a respectively derived method for mapping datasets into networks. The coincidence similarity has been found to allow some specific interesting properties which have allowed enhanced performance (selectivity and sensitivity) concerning several pattern recognition tasks. Several combinations of 20 morphological features were considered, and the respective networks were obtained by maximizing the literal modularity (in supervised manner) respectively to the involved parameters. Well-separated groups were obtained that provide a rich representation of the main similarity interrelationships between the 735 considered neuronal cells. A sequence of network configurations illustrating the progressive merging between cells and groups was also obtained by varying one of the coincidence parameters.