A ridge-based framework for segmentation of 3D electron microscopy datasets

Three-dimensional (3D) electron microscopy (EM) has become a major player in structural cell biology as it enables the analysis of subcellular architecture at an unprecedented level of detail. Interpretation of the resulting 3D volumes strongly depends on segmentation, which consists in decomposing the volume into their structural components. The computational approaches proposed so far have not turned out to be of general applicability. Thus, manual segmentation still remains a prevalent method. Here, a new computational framework for segmentation of 3D EM datasets is introduced. It relies on detection and characterization of ridges (i.e. local maxima). The detected ridges are modelled as asymmetric Gaussian functions whose parameters constitute ridge descriptors. This local information is then used to cluster the ridges, which leads to the ultimate segmentation. In this work we focus on membranes and locally planar structures in general. The performance of the framework is illustrated with its application to a number of complex 3D datasets and a quantitative analysis.