基于荧光显微镜图像的神经丝自动跟踪

AUTOMATED NEUROFILAMENTS TRACKING BASED ON FLUORESCENCE MICROSCOPY IMAGES

神经丝是一种长而柔软的蛋白质有机物,它能在神经细胞中沿着神经轴突快速且随机地运动。神经丝的运动研究对于如神经退行性疾病的诊断等应用是非常重要的。传统的方法在很大程度上依赖于在荧光显微镜图像下手工标记神经丝。这种人工跟踪不但对大量图像实现起来非常费时,而且会带来很多人为跟踪误差。提出神经丝全自动跟踪的方法:粒子滤波跟踪算法和检测跟踪算法。在这两种算法中,都利用了神经丝在神经轴突内运动这一特征。在粒子滤波算法中,限制了粒子的位置和方向,从而显著地降低了粒子使用的数量,大大减少了算法的运算时间。在检测跟踪算法中,提取出沿轴突运动的神经丝轨迹并描绘成一条参数化曲线,利用马尔可夫随机场图形标签来确定包含运动神经丝的轴突块,将神经丝的首端和尾端位置细化到亚像素精度。在实际的实时跟踪实验中,将粒子滤波跟踪算法与检测跟踪算法进行比较,显示出了检测跟踪算法在运算速度和跟踪准确性方面优于粒子滤波算法。

Neurofilaments are the long and flexible protein organic polymers which can move rapidly but randomly along the neural axon of nerve cells.Studying the movement of neurofilaments is important to the applications such as diagnosing the neurodegenerative diseases. Traditional methods rely to a large extent on manual labelling the neurofilaments on fluorescence microscopy images.Such manual tracking is very time-consuming in its implementation on a large number of images,and will bring quite a few human tracking errors.In the paper we present two fully automated neurofilaments tracking method：the particle filtering tracking algorithm and the detection tracking algorithm.In both two algorithms we all make use of the feature of neurofilament moving within the axon.In particle filtering algorithm,we confine the location and orientation of particles so that the numbers of particles used are significantly reduced and the operation time of the algorithm is greatly decreased.In detection algorithm,we extract the trajectory of neurofilaments moving along the axon and depict it to a parameterised curve,and then use graphic label of Markov random field to determine the axon blocks encompassing moving neurofilaments,and refine head-end and tail-end positions of neurofilament to the precison of sub-pixel.In practical real-time tracking experiment,we compare the particle filtering tracking algorithm with detection tracking algorithm,it is showed that the latter outperforms the former in operation speed and tracking precision.