An Adaptive Ray-Shooting Model for Terminations Detection: Applications in Neuron and Retinal Blood Vessel Images

Abstract

2D and 3D termination points are very good seeding point choices for the tree-like structure reconstruction in neuron or retinal blood vessel images. Previously, a ray-shooting model was proposed to detect the termination points in fluorescence microscopy images of neurons, by analyzing the pixel intensity distribution of the neighborhood around the neuron termination candidates. However, the length of the shooting rays and the number of z-slices taken into account in the existing ray-shooting model are fixed empirical number. This ray-shooting model cannot handle the diameter variation of neuron branches. In this paper, we propose an adaptive ray-shooting model to detect the terminations of neurons or retinal blood vessels by changing the length of the shooting rays according to their local diameters. The local diameter is estimated by the Multistencils Fast Marching Method (MSFM) in combination with the Rayburst sampling algorithm. We train a support vector machine (SVM) classifier to classify the termination points and non-termination points, by using the pixel intensity distribution features extracted by the adaptive ray-shooting model. Compared with the previous work, the experimental results on multiple neuron datasets and retinal blood vessel datasets show that our method significantly improves the detection accuracy rate by about 10% in challenging datasets.