High density cell tracking with accurate centroid detections and active area-based tracklet clustering

Accurate cell tracking and lineage construction under microscopy has played an important role in analyzing cell migration, mitosis and proliferation. In the last decade, this labor-intensive manual analysis was gradually replaced by automated cell tracking methods; however, they are often limited to cells with certain morphologies or staining. In this paper, we propose a novel hierarchical tracking framework (Hift), which does not have these limitations. To keep the robustness and feasibility with different cell densities, we concluded several cell motion events into different tracking stages, including entry, exit, division, merge, fast motion, etc. And the fusion of global and local information is applied in both detection and tracking modules, to ensure the flexibility and expansibility of cell detection. To get a full-time cell lineage, we first introduce a conservative distance limit to obtain tracklets with high reliability in the tracking stage. Then motion events are recognized with local information for further corrections. At last, trajectories are linked and completed based on an active search area estimated by the established tracklets. The hierarchical framework designed in Hift enhances the ability of cell detection and tracking by combining the global assignment and local optimization in spatial-temporal dimension. Experimental results of Hift on three large-scale datasets with high cell densities and four sparse datasets demonstrate its efficacy. Hift is available at: http://www.csbio.sjtu.edu.cn/bioinf/Hift/.