| Thanks to the rapid development of microscopy system technology and its widespread propagation, such as the innovative Total Internal Reflection Fluorescence Microscopy (TIRFM), researchers are able to probe and record motion and status transformation of nano-scale vesicles in live cells. Afterwards, image processing methods can be used to analyze the dynamic processes, and as a result it does help to better understand the biological and physiological phenomena. For example, processing images of GLUT4 (Glucose Transporter 4) vesicles, which represent transportation mechanism of Glucose, will promote the solving of diabetes which has been harassing people to some degree.In this research, it is intended to discuss the methods for processing sequential images of thousands of vesicles moving in live cells. Firstly, it is essential to correct the fluorescence bleaching of time-lapse images. In this research, it has utilized the Robust Least-squares combined with Bayesian Estimation to compensate the lost information in the hundreds of sequential images. In fact it is a pre-processing for the further processing and there is some innovative meaning in it. Secondly, this research has discussed a method to detect fluorescence vesicles and track their motions, namely the feature point tracking method based on graphic theory. With the obtained locations of vesicles, their own distribution information can be reconstructed from Gaussian Point Spread Function. And subsequently, in accordance with the optimized linkage rules in local neighborhood, the motions of vesicles can be tracked, and the 3D trajectories can be composed by the aforesaid x-y plane trajectories and the Z-axis relative coordinates which can be obtained from the variation of fluorescence. Consequentially, some physical parameters can be computed from the trajectories. Moreover, this research has emphasized another algorithm for identifying the fusion events and locations automatically based on designed statistics and templates. Actually, nowadays, researchers dominantly detect and record fusions visually, or use Fusion Assistant which has focused too much on details. At last, this research has elucidated a Gaussian Mixture Model to reconstruct every single vesicle from the clusters computed with EM algorithm (Expectation-Maximum).With application to the experimental data, the above stated methods all have produced satisfying results, which are redound to the work of biological or physiological researchers. |
