Cell Tracking Based on Integer Linear Programming and Probability Scores

The cell tracking problem can be formulated as a global optimization problem based on integer linear (IL) programming. The obtained solution depends on the definition of the IL program and the chosen costs of the events considered in the problem formulation. In general, balancing meaningful costs across different events is difficult. To simplify this problem, we propose a cost assignment approach based on probability scores, allowing us to easily interpret the free parameters. The idea is presented based on the BF-C2DL-HSC dataset from the Cell Tracking Challenge. We show our method reaches state-of-the-art performance when evaluating both technical and biological tracking measures.