Tree2Tree: Neuron segmentation for generation of neuronal morphology

The knowledge of the structure and morphology of neurons is a central part of our understanding of the brain. There have been concerted efforts in recent years to develop libraries of neuronal structures that can be used for multiple purposes including modeling the brain connectivity and understanding how cellular structure regulates function. However, at present, tracing neuronal structures from microscopy images of neurons is very time consuming and somewhat subjective and therefore not practical for the current datasets. Current automatic state of the art algorithms for neuron tracing fail to work in neuron images which have low contrast, amorphous filament boundaries, branches, and clutter. In this paper, we develop Tree2Tree, a robust automatic neuron segmentation and morphology generation algorithm. It uses a local medial tree generation strategy for visible parts of the neuron and then uses a global tree linking approach to build a maximum likelihood global tree by combining the local trees. Tests on cluttered confocal microscopy images of Drosophila neurons give results that correspond to ground truth within ±5.3 pixel RMSE margin of error.