Analysis of variable-order interacting multiple model algorithms for cell tracking

In this paper we propose a modification of the Interacting Multiple Model (IMM) filter to effectively track complex dynamics in cell images. Our solution proposes a more efficient use and combination of the multiple Kalman filter estimations that lead to a performance improvement in multi-cell sequences, with an increase of up to 10% in the recall value, compared to the classic IMM. First and second order models are evaluated in the scope of cell migration. The system is evaluated and compared against a baseline using 3D synthetic confocal microscopy images, where cells behave realistically according to actual cell trajectories extracted from real sequences in biology.