Automatic Optimization Method for Segmentation and Surface Model Generation in Electron Tomography

Abstract

Electron tomography can be used to make grayscale volume reconstructions of tissue sections. Images derived from the reconstructions provide the best 3-D spatial resolution currently available for determining the sizes, shapes, and relationships of cellular organelles and macromolecules in situ. Structures of interest are typically examined by segmenting them from the volume and rendering them as surface models according to grayscale values. The fidelity of segmentations and their surface models to the grayscale reconstruction depend on the signal-to-noise ratio (SNR) which can vary considerably between different structures. Current methods of high-fidelity segmentations require tedious manual adjustments. Here, we introduce an automatic optimization method that reduces the manual adjustments, increases the SNR, and improves the fidelity of segmentations and surface models. The method is validated using a well-studied macromolecular assembly in the reconstructions of tissue sections from neuromuscular junctions.