Three-dimensional Particle Tracking Based On Multiple Focal Plane Fluorescence Microscopic Images

In recent years, the green fluorescent protein is usually used as report gene in biomedical area to mark the biological particles or issues and image them. Recognizing and tracking these active biological particles will lead to obtain some movement parameters such as movement velocity and acceleration to further research their dynamic characteristic. Particle dynamics based on tracking fluorescent microscopic particles provides strong research method on some researches like exploring the essence of life, the mechanism of neural activity and the happening of cancer.Limited by the photo-bleaching of fluorescent protein and microscopic optic imaging, Dynamic wide-field fluorescence microscopy images were unable to obtain the frame rate and SNR required by the traditional tracking algorithms. It is the difficult and hotspot issues for tracking the intensive small-scale particles, especially those sheltered and non-smooth moving particles. Making sure of enough frame rate in time domain to satisfy long time tracking, we recover the images through three-dimensional de-convolution applied on multiple focal plane images of less layer, reliably segment particles and finally realize particle tracking and accurately capture the particle fusion and division events. Our research will largely promote the important discoveries of life science research to better serve the national economy and human life.This paper realizes three dimensional particle tracking algorithm based on multiple focal plane (MFPT) and researches the entire framework for particle tracking in fluorescent microscopic images. Specially, three aspects are included:(1) Object recognition algorithm based on the maximum likelihood de-convolution and the undecimated wavelet (MLDC-UWT);(2) Particle filtering Tracking based on corresponding points matching model (CPMM-PF);(3) Particle recognition algorithm based on interpolation and maximum likelihood de-convolution (IML-UWT) for less focal plane images.Firstly, the simulated multiple focal plane images are sampled to form three dimensional sample space and we propose the MFPT algorithm to identify and track the particles in this sample space. Separately the MLDC-UWT recognition algorithm is proposed to implement the recognizing process, that is, the maximum likelihood de-convolution algorithm (MLDC) is involved to remove the noise during the image acquisition and the effect from particles in non-focal planes. Then the a trous wavelet transform (UWT) is introduced to consider the multi-scale product of wavelet images after filtering the non-significant coefficients to enhance the rough area of target particles. The potential targets are further accurately ensured through the method of calculating the local intensity maximum for each dimension (ED-LIM).Second, as the limited features of particles in fluorescence microscopy images, the position and intensity of particles is the two chosen characteristics, and Euclidean distance is chosen to be the cost function of the corresponding points matching model (CPMM). The particle filtering (PF) algorithm is used to realize particle tracking.Finally, the sample interval must be considered as short as possible when sampling the focal plane images from the actual observed biological sliders, so the recognition algorithm based on interpolation and de-convolution (IMLDC-UWT) is discussed in less layers of focal plane image samples. And our method also has relatively high recognition rate by padding zeros in frequency domain to achieve the expanded layers of samples and using maximum likelihood de-convolution algorithm and the undecimated wavelet transform, compared with single focal plane UWT recognition algorithm (sUWT) and multiple focal plane UWT recognition algorithm (mUWT).