Fast reconstruction of water-tight surface mesh of neurons

Neuron morphology reconstruction from high-resolution imaging data is essential for understanding the structure and function of the brain in neuroscience. However, previous methods cannot achieve both water-tight and high performance in surface mesh reconstruction of large-scale neurons. Thus, this paper proposes a novel neuronal surface mesh reconstruction algorithm based on isosurface extraction, virtual memory management, and parallel computation. The space of a neuron is firstly divided into blocks, and they are organized as a sparse octree to handle large-scale neurons with long projection. We then perform voxelization and isosurface extraction on valid blocks based on the skeleton model of the neuron to ensure the generated mesh that is water-tightness, and the quality and the density of the mesh are controllable. Since each block is processed independently, the reconstruction can be performed in parallel for high performance and partially for interactive modification during neuron proofreading. Experiments demonstrate that the proposed algorithm can generate water-tight neuronal surface meshes effectively and satisfy the needs of interactive visualization and correction.