CONSIDERATIONS FOR LOW-DOSE IMAGING IN TRANSMISSION ELECTRON MICROSCOPY

Abstract

Transmission electron microscopy (TEM) plays an essential role in material science and biology. After the substantial improvement of electron detectors around 2012, radiation damage caused by electron-sample interaction becomes the primary factor limiting image quality. Although the newly developed direct electron detectors can count single-electron events and record high-frame-rate movies, these features have not been fully exploited. This thesis explores probabilistic models for image registration under low-dose conditions. Specifically, we have formulated a Poisson likelihood-based motion correction algorithm for low-dose images and found that electron dose fractionation over a various number of images is an optimizable factor in terms of image registration. For beam-induced damage, we have established a toy model that aims to infer damage quantified by B-factors from observed Poisson counts. The relevant analysis shows the potential benefit of low-dose imaging, which provides an indirect way to alleviate the effect of radiation damage in electron microscopy.