A random-sampling approach to track cell divisions in time-lapse fluorescence microscopy

• Main Authors: Saoirse Amarteifio, Todd Fallesen, Gunnar Pruessner, Giovanni Sena
• Format: Article
• Language: English
• Published: BMC 2021-03-01
• Series: Plant Methods
• Subjects: Plant development; Plant root; Arabidopsis; Light-sheet microscopy; CYCB::GFP; Transient fluorescence
• Online Access: https://doi.org/10.1186/s13007-021-00723-8

Abstract Background Particle-tracking in 3D is an indispensable computational tool to extract critical information on dynamical processes from raw time-lapse imaging. This is particularly true with in vivo time-lapse fluorescence imaging in cell and developmental biology, where complex dynamics are observed at high temporal resolution. Common tracking algorithms used with time-lapse data in fluorescence microscopy typically assume a continuous signal where background, recognisable keypoints and independently moving objects of interest are permanently visible. Under these conditions, simple registration and identity management algorithms can track the objects of interest over time. In contrast, here we consider the case of transient signals and objects whose movements are constrained within a tissue, where standard algorithms fail to provide robust tracking. Results To optimize 3D tracking in these conditions, we propose the merging of registration and tracking tasks into a registration algorithm that uses random sampling to solve the identity management problem. We describe the design and application of such an algorithm, illustrated in the domain of plant biology, and make it available as an open-source software implementation. The algorithm is tested on mitotic events in 4D data-sets obtained with light-sheet fluorescence microscopy on growing Arabidopsis thaliana roots expressing CYCB::GFP. We validate the method by comparing the algorithm performance against both surrogate data and manual tracking. Conclusion This method fills a gap in existing tracking techniques, following mitotic events in challenging data-sets using transient fluorescent markers in unregistered images.