Tracking of motile bacteria with instance segmentation aided by semi-synthetic image augmentation and quantitative analysis of run-and-tumble motion

Abstract

Motile bacteria represent a paradigmatic class of living active matter, attracting interest across disciplines ranging from physics and biology to small-scale robotics. While various tracking approaches have been developed, resolving individual cells in contact has been relatively underexplored despite its relevance to the analysis of collective motion. Here, we present a tracking pipeline that distinguishes partially overlapped bacterial cells using embedding-based instance segmentation trained solely on semi-synthetically augmented images, eliminating the need for manual labeling. The trained network performs reliably in both wide-separation and in-contact scenarios, demonstrating potential for single-cell tracking even in frequently colliding or moderately dense environments. The semi-synthetic dataset also proves effective for training another tracking algorithm, although the algorithm fails to resolve in-contact scenarios at a comparable level. As an application, we analyzed the extracted trajectories using a stochastic model of bacterial swimming based on run-and-tumble dynamics. This model incorporates Cauchy noise to describe abrupt angular reorientations and enables the quantification of how swimming behavior systematically varies with temperature. This quantification framework illustrates a general approach for linking observed motility to underlying behavioral parameters under controlled conditions.