Automatic Cell Cycle Analysis Based on Live Cell Fluorescence Microscopy Image Sequences

In this thesis, we address the problem of automatic analysis of large sets of 2D and 3D fluorescence microscopy image sequences. Increasingly automated laboratories allow biologists to perform large-scale screening experiments which easily produce TBytes of image data. However, analyzing this data mass is often problematic since manual analysis is not feasible, and suitable automated image analysis approaches are often lacking. Here, we present an automatic approach
for detailed cell cycle analysis based on live cell fluorescence microscopy image sequences.

In particular, our approach enables robust automatic determination of cell cycle phase durations which allows automatically analyzing large-scale screening experiments to study mitotic gene regulation. To this end, we developed a complex image analysis workflow comprising of methods for segmentation, tracking, feature extraction, classification into cell cycle phases, and phase sequence parsing. Our approach was applied to a large number of real images from four different screens. In total, we analyzed more than 1000 image sequences with around 100 to 200 time steps each. The experimental evaluation showed that our approach robustly and accurately solves different tasks: (1) classification of cell nuclei into different cell cycle phases, (2) determination of cell cycle phase durations, and (3) classification and quantification of morphological phenotype classes. In our experiments we also demonstrated the applicability of our approach for different biological settings by analyzing images of different cell lines and images of different screening microscopes.