Automated identification of microtubules in cellular electron tomography

We describe a method for automatically finding the location and conformations of microtubules in tomograms of high-pressure frozen, freeze substituted cells. Our approach uses two steps: a preprocessing step that finds locations in the tomograms that are likely to lie inside microtubules and a tracking step that uses the preprocessed data to identify the trajectories of individual microtubules. We test this method on a reconstruction of a Caenorhabditis elegans mitotic spindle and we compare our results with those obtained by a human expert who manually segmented the same data. At present, the method could be used to assist the analysis of large-scale tomography reconstructions. With further improvements, it may be possible to reliably segment cellular tomograms without human intervention.