Abstract—Quantitative analysis of cell dynamic processes
through fluorescence microscopy imaging requires
simultaneously tracking large and time-varying number of
bright spots and its individual states in noisy image sequences.
Such process is characterized as a challenging task due to
several roadblocks including the severe image noise and clutter,
the occlusion of one cell by others, and the weak image contrast.
In this paper, we propose a novel ant stochastic searching
behavior based tracking algorithm, which is called ANT
estimator, to tracking multiple cells in fluorescence image
sequences. In our ant system, each ant determines
probabilistically potential state and then adjusts its mobility
according to cell detection position heuristic information.
Simulation results verify the effectiveness of our algorithm
when applied to cell tracking cases, and its performance is also
compared with the particle filter based cell tracking algorithm.
I. I
NTRODUCTION
H
E
ability to visualize cells dynamic processes in space
and time has been made possible by revolutionary
developments in imaging technology in the past two decades.
In terms of hardware, advances in optical systems design
have taken light microscopy from widefield to confocal and
spinning disk microscopy [1-3], and this in turn results in
vast amount of dynamic image data. With large time span
and more cells being concerned, data exploration is inevitable,
and general methods of processing image data do not
guarantee sufficient speed, accuracy, and reproducibility.
Therefore, many efforts have been made to develop novel and
effective cell tracking techniques.
Generally speaking, previous work on cell tracking can be
classified into deterministic tracking and stochastic tracking
[4]. Deterministic tracking usually relies on accurate and
reliable segmentation results, and sometimes yields poor
performance for severe image quality. Meanwhile, stochastic
tracking is usually characterized as a parameter estimation
problem, in which a stochastic state transition model and
observation model between the target state and the image
intensity observation are first built, and then appropriate
multi-target algorithms [5-8] are required to approximate the
evolution of the target state. It is recognized that the
stochastic tracking is a preferred automated tracking method,
since it can replace tedious manual procedures and eliminate
the bias and variability in human judgments. Particle Filter
(PF) is a widely used non-linear non-Gaussian parameter
filtering algorithm, and the use of PF in combination with
Benlian Xu is with the Changshu Institute of Technology, Changshu, P. R.
China, 215500 (e-mail:xu_benlian@ yahoo.com.cn).
level-sets [9] and active contours [10] has been reported and
achieved better performance than deterministic methods for
biological cell [11].
Ant Colony Optimization (ACO) makes use of simple agents
called ants which iteratively build candidate solutions to a
combinatorial optimization problem, and the ants’ solution
construction is guided by pheromone trails and
problem-dependent heuristic information. It is observed that
generic ACO algorithm could be successfully applied to any
combinatorial optimization problem [12], and its variants to
parameter estimation field as well [13,14]. Inspired by ant
stochastic searching behavior, we develop a suitable ANT
Estimator, a colony of moving ants each represented by
individual state, for tracking simultaneously large number of
cells in 2-D dynamic fluorescence microscopy images.
The paper is organized as follows. In Section II, we first
describe the background of PF, and then give the idea origin
of ant estimator . In Section III, the ant estimator for multiple
cells is designed and analysized as well. Simulation results
are given for demonstrating the superior performance of our
algorithm in Section IV.
II. B
ACKGROUND
A. Particle Filter and It’s Application to Cell Tracking
In general, the state of each cell evolves in a non-linear way,
and this case is just suitable for the application of particle
filter. We assume that the dynamic model of each cell is
described as
1
t t t
f
h
(1)
where
t
x
denotes the state at time
,
t
z
is the
corresponding measurement,
and
represent the process
and measurement noise, respectively. In the above formula,
the system model is assumed to follow the first order Markov
density function
t t t t
p
on the state space
R
,
and the observation function
is usually modeled as a
likelihood function
p
on the observation space
, respectively.
Given an initial density function
of state, the
posterior filtering density at time
can be calculated by the
Bayesian recursion
| 1 1: 1 | 1 1| 1 1: 1
t t t t t t t t t t
(2)
| 1 1: 1
| 1:
| 1 1: 1
( | )
t t t t t t t
t t t t
t t t t t
p
z x x z
x z
(3)
Multiple Cell Tracking Using Ant Estimator
Benlian Xu, Mingli Lu, Peiyi Zhu, Qinglan Chen, and Xiaoying Wang
2012 International Conference on Control, Automation and Information Sciences (ICCAIS) MA01
978-1-4673-0813-7/12/$31.00 ©2012 IEEE 13