3D Neuron Branch Points Detection in Microscopy Images

Abstract

Neuron tracing (reconstruction) is an important step toward understanding the functionality of neuronal networks. Neuron termination points and branch points, collectively called critical points, play an important role in neuron tracing applications. There are some existing methods for 3D neuron termination points detection. However, 3D branch points detection method has barely been explored. In this paper, we propose a 3D branch points detection method in microscopy images by reverse-mapping the 2D branch points back into the 3D space, according to the pixel intensity distribution along the projection direction. The 2D branch points are detected by an adaptive ray-shooting model in 2D maximum intensity projections (MIPs), where the center is the 3D branch point candidates, of a specified number of adjacent slices along the Z direction. The adaptive ray-shooting model analyzes the intensity distribution of the neighborhood around the branch point candidates and is robust to neurite diameter variations. The experimental results on multiple neuron image datasets show that our proposed method can achieve an average false negative rate and false positive rate of 15.67% and 10.67% for neuron branch point, respectively.