Bayesian image analysis with on-line confidence estimates and its application to microtubule tracking

Abstract

Automated analysis of fluorescence microscopy data relies on robust segmentation and tracking algorithms for sub-cellular structures in order to generate quantitative results. The accuracy of the image processing results is, however, frequently unknown or determined a priori on synthetic benchmark data. We present a particle filter framework based on Markov Chain Monte Carlo methods and adaptive annealing. Our algorithm provides on-line per-frame estimates of the detection and tracking confidence at run time. We validate the accuracy of the estimates and apply the algorithm to tracking microtubules in mitotic yeast cells. This is based on a likelihood function that accounts for the dominant noise sources in the imaging equipment. The confidence estimates provided by the present algorithm allow on-line control of the detection and tracking quality.