Fast Marching Spanning Tree: An Automatic Neuron Reconstruction Method

Abstract

Neuron reconstruction is an important technique in computational neuroscience. There are many neuron reconstruction algorithms, but few can generate robust result, especially when a 3D microscopic image has low single-to-noise ratio. In this paper we propose a neuron reconstruction algorithm called fast marching spanning tree (FMST), which is based on minimum spanning tree method (MST) and can improve the performance of MST. The contributions of the proposed method are as follows. Firstly, the Euclidean distance weights of edges in MST is improved to be more reasonable. Secondly, the strategy of pruning nodes is updated. Thirdly, separate branches can be merged for broken neurons. FMST and several other reconstruction methods were implemented on the 120 confocal images of single neurons in the Drosophila brain downloaded from the flycircuit database. The performance of FMST is better than some existing methods for some neurons. So it is a potentially practicable neuron construction algorithm. But its performance on some neurons is not good enough and the proposed method still needs to be improved further.