A Highly Accurate Model Based Registration Method for FIB-SEM Images of Neurons

Abstract

Focused Ion Beam Scanning Electron Microscope (FIB-SEM) imaging is a technique that image materials section-by-section at nano-resolution, e.g.,5 nanometer width voxels. FIB-SEM is well suited for imaging ultrastructures in cells. Unfortunately, typical setups will introduce a slight sub-pixel translation from section to section typically referred to as drift. Over multiple sections, drift compound to skew distance measures and geometric structures significantly from the pre-imaged stage. Popular correction approaches often involve standard image registration methods available in packages such as ImageJ or similar software. These methods transform the images to maximize the similarity between consecutive two-dimensional sections under some measure. We show how these standard approaches will both significantly underestimate the drift, as well as producing biased corrections as they tend to align the images such that the normal of planar biological structures are perpendicular to the sectioning direction causing poor or incorrect correction of the images. In this paper, we present a highly accurate correction method for estimating drift in isotropic electron microscope images with visible vesicles.