NRRS: A re-tracing strategy to refine neuron reconstruction

Abstract

Motivation: A fine digital tracing of complete neural morphology from a whole brain imaging dataset is highly required in the neuroscience field. With the participation of human experts in the reconstruction workflow, the complicated annotation challenge brought by the whole brain scale could be addressed. Yet different degrees of deviation at branches and bifurcation point are produced, which hurts the utility of the reconstruction data in downstream analysis. Results: We propose a neuron reconstruction refinement strategy (NRRS) to fill the gap mentioned ahead. Our strategy partitions neuron morphology reconstruction into fixed-size segments and resolves the deviation problems by two steps of neuron tracing. Furthermore, we build a synthetic dataset severed as validation data for quantifying the exact performance. We show that NRRS can handle most of the defined deviation errors while existing solutions hardly achieve matched performance. In applying our method to SEU-Allen dataset containing nearly 2000 complete neuron reconstructions, we present large improvements obtained in the accuracy of the neuron skeleton representation, the task of radius estimation and axonal bouton detection. The results demonstrate the indispensability of our strategy in refining neuron morphology reconstruction.