Retrieving Similar Substructures on 3D Neuron Reconstructions

Abstract

An accurate neuron reconstruction is very important to understand neuron morphology and function, but it is still a challenging task due to the time consuming of manual tracing and the unsatisfactory accuracy of automatic tracing. One way to address the challenge is generating a reconstruction automatically and then checking and amending the result manually. Aiming at implementing this process efficiently, we propose a pipeline to retrieve substructures on one or more neuron reconstructions, which are very similar to a marked problematic substructure and need to be checked one by one. The pipeline consists of four steps: getting a marked substructure, constructing a query substructure, generating candidate substructures and retrieving most similar substructures. The retrieval procedure was tested on 163 gold standard reconstructions provided by the BigNeuron project and a reconstruction of a mouse’s large neuron. Experimental results showed that the implementation of the proposed methods is very efficient and all retrieved substructures are very similar to the marked one in numbers of nodes and branches, and degree of curvature.

Smart imaging to empower brain-wide neuroscience at single-cell levels

“…Following the neuron tracing, to comprehensively analyze the traced morphometry data is critical to unravel the spatial properties of neurons and networks at multiple scales and to understand the mechanisms behind the nervous systems [ 143 – 145 ]. Many techniques have been developed for this aim [ 146 , 147 ]. Among those the morphological grouping has been vastly applied, with the support of many similarity analysis [ 148 – 150 ] and clustering methods, such as UMAP [ 151 ], K-Means [ 152 ], and HCA [ 153 ].…”

Large scale similarity search across digital reconstructions of neural morphology

“…A different method for global search employed an asymmetric binary coding strategy based on the maximum inner product [15], while encoding of morphology with hashing forests has shown promising performance on large datasets [16]. Another variant is to query sub-structures of the neuron with graph representations of the (sub-)trees [17] or using structure tensors and expanding this field via gradient vector flow [18]. A conceptually related problem is to compare two morphological reconstructions of one and the same neuron, such as when benchmarking an automated tracing software against the gold standard of expert manual proofreading [19].…”