Detection of 3D filamentous networks from tomographic electron microscopy

This paper presents a computational approach for detection of filamentous networks in 3D tomographic electron microscopy. Due to the general heterogeneity of chemical staining, imaged signatures may appear punctate and discontinuous. Very often, there is a necessity to characterize organization and phenotypic signatures of stained structures. This is the example of polysaccharides in the plant cell wall, where characterization of its inner structure allows for studies of their mechanical properties and accessibility. Our approach consists of three steps: (i) Hessian filtering for signal enhancement; (ii) tensor voting for detection and completion of organized patterns; and (iii) curve tracking for segmentation of filaments and junctions. We validated our approach by detecting filaments in synthetic images and real plant cell walls imaged by 3D tomographic electron microscopy.