A Spatio-Temporal Algorithm for Resolving Time-Lapse Imaging of Bacteria

With recent advances in microscopy and in-vivo fluorescent labelling, an increasing number of biological studies use time-lapse imaging to quantitatively follow dynamics in living organisms. These studies vary in scale from single molecules, their segregation during cell division to the evolution of cells in an organism or a tissue (embryogenesis, cancer evolution, etc.). One such study, led by the INSERM/Descartes team, aims at understanding the phenotypic variability within lineages of clonal bacteria by quantifying and correlating bacterial growth rate with specific transcription, translation or DNA markers. Two types of images are acquired: phase-contrast to follow the bacteria and fluorescence to follow the markers of interest. Currently the image processing is semi-automatic and its duration (2-3 days per 100 frames) sets a serious limit. Indeed, classical image analysis techniques based on two-step approaches (segmentation, then tracking) reach their limits when the segmentation step cannot be performed on individual images in a reliable way, thus requiring costly human post-processing. Hence, to increase the efficiency of the automated image processing, we propose to perform segmentation and tracking simultaneously, thus aiming at resolving segmentation ambiguities using the temporal redundancy of the data.