Time lapse microscopy of living cells is essential in the assessment of their physiological function. Due to the technical progress image data sets become bigger and exceed the possibility of manual segmentation. Many potent imaging algorithms already exist, however it is a task on its own to adjust them to specific data. While life science researchers often have specific ideas about the distinct features of interest, the same researches do often not know how to convert their thoughts into machine readable form. We address a gap between computer scientists and biomedical researchers by an algorithm, which is trained solely by the expertise of a life science researcher. Thus, it is the aim of this study to construct a trainable image segmentation algorithm enabling life science researchers to convert their ideas to algorithms without coding. We apply our approach to the example of a wound healing assay (HUVEC cells were observed over 24 h with 6 frames per hour resulting in 145 images) to assess collective cell motility quantitatively. To assess the so-called cell covered area over time a three step procedure was established. First the experimenter used a paintbrush like tool to specify cell covered areas within a single frame of the image series. In a second step Bayesian parameter estimation based on the nested sampling algorithm was performed to automatically adjust all parameters of a corresponding image processing pipeline, which segments the cell covered areas. The imaging pipeline consisted of individual parameter dependent imaging functions such as edge detection, blurring or intensity as well as particle size thresholding. The Bayesian parameter estimation is based on a distance metric allowing quantifying the difference of the manually segmented image and the result of the pipeline. Thus, the trained parameter set of the optimized pipeline was obtained by the algorithm as well as their uncertainties. In the final step the imaging pipeline was applied to all frames of the data set reducing the time expenditure for user input from 145 to a single frame. We applied the algorithm to many different images recorded using bright-field, phase contrast or differential interference contrast microscopy to check its stability. In all cases, even for bad image qualities, we obtained reliable results. In addition we could show that different prior knowledge of the experimenter is considered by the algorithm: The input of four persons with differences in accuracy of the manual segmentation caused uncertainties of the border of the cells between 6.8 ± 0.1 pixels and 37.9 ± 0.5 pixels. In conclusion, we present a new robust algorithm allowing automated segmentation of bio-medical image sequences. Our approach is highly modular, since user input, image processing pipelines and metrics are interchangeable if necessary for other problems. We believe that this tool will help to overcome the increasing gap between computer scientists and biological researches.