Study And Application Of Computer Vision For Acquisition Of Multi-dimensional Information Of Cells

Cells as the fundamental units of life structure and function changes along with life processes, including cell growth, cell development, cell metabolism, cell aging and death. Acquisition and study of multi-dimensional information of cells are significant for understanding the essential mechanism of life. As a powerful tool to acquire information at micro-nano scale, an atomic force microscope(AFM) can be utilized to scan the surface of sample, manipulate nano-particles and detect physical mechanical information such as Young modulus, viscosity and molecular force, of cells through AFM contact force. In these operations, the optical microscopy integrated in the AFM system is mainly used to acquire optical images of samples and guide AFM tip to position. However, there exist some issues for the operations with the optical microscopy. The optical images of samples lack of effective and efficient analysis method. On the other hand, the operation for AFM tip positioning through the optical microscopy is manual, which causes tip positioning inaccurate and inefficient. In order to solve these two problems, in this dissertation, an automated system is developed for cellular recognition and extraction of geometric features of cells from optical images of cell patterns and an depth estimation approach from the optical defocus images of AFM probe is studied for automatic AFM tip engagement and positioning. The works are detailed as follows:(1) Automated system for recognizing and measuring a large number of cells which are cultured on the substrate patterned using Optically-induced dielectrophoresis(ODEP), an efficient method to guide cells to grow and form the desired structure, is developed. Active contour model, improved by simplifying initial condition and dynamic programming, is used to recognize the contours of cells, from which geometric features of cells are extracted. Finally, the relationship between the desired structure of substrate and geometric features of cells is analyzed.(2) Blur edges of images are used to estimate the depth of edge from 2D optical images. At first, Gaussian filter is used to smooth the noise of original images. Then the blur size of images is estimated through the ratio between smoothed image and re-blurred image in which the parameters are known. According to the thin lens model, the edge depth of image map can be obtained from the blur size. Finally, experiments with Hi Rox-7700 have verified the validity and feasibility of this approach and the edge depth of microscope image can be estimated in high resolution.(3) According to the special geometric structure of AFM cantilever, two features including edge feature and control point feature of AFM cantilever are extracted. With these two features, the location of the AFM cantilever appearing on the optical image is estimated. Then the cantilever deflection during the engagement process of AFM probe can be calculated through the image registration method. Finally, experiment with the Catalyst AFM is utilized to verify the validity and accuracy of the method.Overall, this dissertation focuses on the study and application of computer vision for multi-dimensional information acquisition of cells. Firstly an automated system for recognition and measurement of cell growing on substrate pattered by ODEP using PEGDA hydrogel is developed. Secondly, a method for depth estimation based on blur size of edge is studies. At last, based on the edge features and control point features, tan image registration method to estimate the error during engagement process of AFM is proposed and verified by the experiment on Catalyst AFM. These work will improve the effectiveness and efficiency of the operations including acquisition and analysis of cellular geometric information and AFM tip engagement and positioning..