Image Segmentation of Bacterial Cells in Biofilms
Bacterial biofilms are three-dimensional cell communities that live embedded in a self-produced extracellular matrix. Due to the protective properties of the dense coexistence of microorganisms, single bacteria inside the communities are hard to eradicate by antibacterial agents and bacteriophages....
|PDF Full Text
No Tags, Be the first to tag this record!
|Bacterial biofilms are three-dimensional cell communities that live embedded in a self-produced extracellular matrix. Due to the protective properties of the dense coexistence of microorganisms, single bacteria inside the communities are hard to eradicate by antibacterial agents and bacteriophages. This increased resilience gives rise to severe problems in medical and technological settings. To fight the bacterial cells, an in-detail understanding of the underlying mechanisms of biofilm formation and development is required. Due to spatio-temporal variances in environmental conditions inside a single biofilm, the mechanisms can only be investigated by probing single-cells at different locations over time. Currently, the mechanistic information is primarily encoded in volumetric image data gathered with confocal fluorescence microscopy. To quantify features of the single-cell behaviour, single objects need to be detected. This identification of objects inside biofilm image data is called segmentation and is a key step for the understanding of the biological processes inside biofilms.
In the first part of this work, a user-friendly computer program is presented which simplifies the analysis of bacterial biofilms. It provides a comprehensive set of tools to segment, analyse, and visualize fluorescent microscopy data without writing a single line of analysis code. This allows for faster feedback loops between experiment and analysis, and allows fast insights into the gathered data.
The single-cell segmentation accuracy of a recent segmentation algorithm is discussed in detail. In this discussion, points for improvements are identified and a new optimized segmentation approach presented. The improved algorithm achieves superior segmentation accuracy on bacterial biofilms when compared to the current state-of-the-art algorithms.
Finally, the possibility of deep learning-based end-to-end segmentation of biofilm data is investigated. A method for the quick generation of training data is presented and the results of two single-cell segmentation approaches for eukaryotic cells are adapted for the segmentation of bacterial biofilm segmentation.