Daniela Franz


A4 Automated and interactive learning image-analysis for fluorescence-microscopy by the example of microbial effectors in host cells

Principal investigator
Thomas Wittenberg

Ralf Palmisano

A wizard-based approach to generic image analysis for fluorescence microscopy images

Fluorescence microscopy is a valuable tool for the study of microbial effectors. In the resulting microscopic images the spatial extent, position, and the related intensities of cells, nuclei, or nuclear dots can be determined. Nevertheless, this requires the evaluation of a large base of image-stacks with one image per stain. The manual evaluation of such data is a time-consuming task. Therefore, only a small subset of the available data is evaluated. Additional to the selection of the data subset, intra- and inter-observer variability leads to varying results. Automated image analysis does not only improve the objectivity and reliability of the results, but also the quantity of the data, that can be evaluated in the given time. The development of a specific image analysis pipeline takes both expertise in image processing and in understanding and interpreting microscopic data. For each kind of image data, a different image analysis pipeline needs to be developed. Computer scientists develop these pipelines in most cases. In a first step, this work considers the development of flexible image analysis pipelines, which facilitate rapid-prototyping for the evaluation of many different microscopic image evaluations. In most cases computer scientists are no experts in understanding and interpreting microscopic images and the experts with microscopic data, such as microbiologists, immunologist, or virologist (application experts), lack image processing knowledge. In a second step, this work considers a wizard-based approach to fluorescence microscopy image analysis. The goal is to develop a tool that guides an application expert through the development of an image analysis pipeline tailored to a specific evaluation of fluorescence microscopic images. Supervised learning is applied to obtain the parameters and the series of the appropriate image analysis methods for evaluation specific image analysis pipelines. In a wizard, all learning and image processing terminology is avoided and instead the image analysis pipeline is built gradually by guiding the user step-by-step through this process and letting him rate the quality of intermediate and final image processing results.

Figure: Example workflow for a wizard-based segmentation of fluorescence microscopy images.





July 2014 6th Annual Retreat, Erlangen School of Molecular Communication, Kloster Banz, Bad Staffelstein, Germany
"User Guidance for Active Contour-based Cell Segmentation"