The Time Series Image Analysis of the HeLa Cell Using Viscous Fluid Registration

Abstract

Optical microscopy image analysis is important in the life science research. To obtain the motion of the cell, we use the viscous fluid registration method based on fluid dynamics. Viscous fluid registration deforms an image at time t to the next image at time t+1. In this algorithm, there is a problem that an object cannot be divided into two. In other words, the divided objects from one are connected by thin line because the velocity field on the connected thin line is zero. To solve this problem, we suggest a new viscous fluid registration algorithm for the object division. This algorithm is only added similarity maximization step to correct the displacement in the near pixels in the original viscous fluid registration. Using this method, one object is divided into two, and divided objects are not connected. We experiment the anaphase detection based on a nucleus identification using laser scanning microscope HeLa cell images. Experimental result shows that 74 in 76 cells are tracking well and 6 cells in the anaphase are detected. In three scenes in the cell division which can not be divided into two using original viscous fluid registration, suggested algorithm can be divided into two cells completely.