Motion Refined Machine Learning Enables Comprehensive Characterization of Bacterial Swarming Dynamics

Bacterial swarming on semi-solid surfaces exemplifies collective motion driven by physical interactions. We investigate the swarm front of Enterobacter sp. SM3, a gut bacterium exhibiting strong swarming behavior. A controlled dilution produces a quasi-two-dimensional monolayer amenable to high-resolution imaging. To overcome the challenge of dense, motile populations, we combine deep-learning segmentation (CellPose) with a PIV-guided Intersection-over-Union (IoU) propagation scheme that preserves cell identity across frames. This enables quantitative single-cell tracking and analysis of collective dynamics. Mean square displacement and velocity correlations reveal sustained superdiffusive motion arising from steric alignment and hydrodynamic coupling, linking microscopic interactions to emergent swarm organization.