gms | German Medical Science

Introduction: The analysis of in vitro cultured (adult) stem cells is an important tool that helps to gain new insights in the wide field of regenerative medicine, i.e. ontogenesis and homeostasis [1]. The most common method for cellular analysis is a fluorescence staining. Unfortunately, it only allows for the extraction of a static snapshot of the protein expression of each single cell. Therefore, the complete dynamic process of an in vitro cell culture, i.e. the relationship and the movement of the cells, can not be assessed or analysed. Here, we propose a novel method based on a geometric hashing approach that allows for the combination of a cellular genealogy with fluorescence staining using time lapse microscopy. This combination will help to correlate the protein expression with the cells relationship, which is a first step towards the analysis and modelling of cell fates in a dynamic, systems biological approach. While it is still challenging to combine the time series from long term observation with fluorescence staining, we believe that this method will help to unravel the regulating mechanisms behind cell proliferation and differentiation.

Material and Methods: In the described setup, time lapse microscopy is used to record a stem cell population over several days at different position. Subsequently, the cells are detected and the cellular genealogy is extracted from the phase contrast (PH) images using a cell tracking approach [2]. However, the exact position of the recorded images gets lost during the following staining procedure. To overcome this problem, we use a fluorescence microscope to image a big area of the cell population. This mosaic image contains the cells that were recorded during the long term observation. Now, the task is to identify the same cells in both datasets, in the fluorescence and the phase contrast image data. For this, we use a geometric hashing based approach to identify the cells in an image by their unique constellation. A similar approach was used to identify stars in astronomical image data before [3].

Results: The introduced algorithms were implemented in MATLAB and evaluated and tested using two data sets. These data sets are comprised of two big FL images (>100 / 300 MegaPixel) showing 2.000 and 10.000 cells (adult stem cells from rat / human) and several phase contrast images (25 / 5) taken from the same cell populations. Each of the PH images shows a small but unknown fraction of the FL image data. We show that our algorithms can determine the exact position of each single image at a high accuracy, even in the presence of a strong segmentation error, i.e. 50% cell detection error.

Discussion: We presented a robust method to fuse microscopical image data from different origin, i.e. we can combine phase contrast, fluorescence and other image modalities based on a geometric hashing approach. Here, the only prerequisite is object segmentation. In the presented context we can combine a cellular genealogy with a protein expression which means a new perspective for the systems biology on single cell level.