Deformable trellises on factor graphs for robust microtubule tracking in clutter

Abstract

A main challenge in microtubule tracking is due to clutter, or the presence of many similar intersecting structures. This paper proposes a two-layered probabilistic formulation which has at its foundation a factor graph serving as a multi-label inference engine designed to provide distinction between open contours of interest and other microtubules or noise. The second layer is a deformable trellis defined over the resulting label probability map, where a Hidden Markov Model (HMM) is employed to determine the most probable current location of the microtubule body. The overall framework enjoys the "best of both worlds" -the factor graph is effective in discriminating between contours of interest and others that exhibit similar statistical properties, while the deformable trellis with its HMM offer accurate modeling of microtubule dynamics in terms of growth and shortening, as well as precise body tracing, accounting for prior information, all within a principled Bayesian framework. Simulation results provide evidence that the proposed approach outperforms existing techniques.

Automated Tracking of Biopolymer Growth and Network Deformation with TSOAX

“…Very little work however exists to quantify the evolution of network structures in 2D/3D time-lapse images. Some previous efforts focused on tracking of individual filaments [Sargin et al, 2011, Kidambi et al, 2012, Smith et al, 2010 or filament tips [Applegate et al, 2011].…”