Summary | Background | Methods | Results | Publications

Model-Based Automated Segmentation
The goal of this research is to improve the accuracy and efficiency of volume segmentation by developing an automated segmentation method. In particular, we investigate how to systematically incorporate prior knowledge to make robust segmentation of difficult electron tomography images. Due to the inherent difficulties in sub-cellular images and the limitations of the existing methods, the successful segmentation of microtubules and the filamentous plus-ends also requires the combination of multiple methods into the segmentation. To this end, we propose a model-based automated approach to extracting the microtubules and the plus-ends with a coarse-to-fine scale scheme consisting of volume enhancement, model-based microtubule segmentation and probabilistic plus-end tracing.
In volume enhancement, we fully exploit the local and global geometric properties of the tubular objects. An invariant and anisotropic wavelet filtering is first presented to globally enhance the elongated structures. This is followed with 3D geometric filters that locally accentuate the tubular objects and diminish objects of other morphologies. For the segmentation of microtubules in the enhanced volume, an improved statistical shape model is first built to capture the shape of the cross-sectional contours of the tubular objects. It is then combined with Kalman filtering to impose smoothness constraint along the longitudinal direction of the tubes. For tracing the finer structures connected with the tubular objects, we propose a probabilistic tracing method by integrating both the shape and appearance knowledge of the curvilinear features into a particle filter framework.

Summary | Background | Methods | Results | Publications