Compression, Segmentation, and Modeling of Filamentary Volumetric Data

Abstract

We present a data structure for the representation of filamentary volumetric data, called the L-block. While the L-block can be used to represent arbitrary volume data sets, it is particularly geared towards representing long, thin, branching structures that prior volumetric representations have difficulty dealing with efficiently. The data structure is designed to allow for easy compression, storage, segmentation, and reconstruction of volumetric data such as scanned neuronal data. By ''polymerizing'' adjacent connected voxels into connected components, L-block construction facilitates real-time data compression and segmentation, as well as subsequent geometric modeling and visualization of embedded objects within the volume data set. We describe its application in the context of reconstruction of brain microstructure at a neuronal level of detail.