U.S. Department of Energy

Pacific Northwest National Laboratory

Quantifying Feature Bias and Uncertainty Derived from Sub-sampled Low-Dose Scanning Transmission Electron Microscopy

Publish Date: 
Tuesday, June 20, 2017
Scanning transmission electron microscopes (STEM) provide high resolution images at an atomic scale. Unfortunately, the level of electron dose required to achieve these high resolution images results in a potentially large amount of specimen damage. A promising approach to mitigate specimen damage is to subsample the specimen [1, 2, 3]. With random sampling, the microscope creates high resolution images of segments of the specimen while reducing overall damage. However, subsampling produces images that have several missing values and can be hard to interpret and analyze in their raw state. In order to make the subsampled images more interpretable, several methods have been proposed to recreate the full images from the subsampled images (in-painting). Most notable among them is compressive sensing, which has been shown to be an extremely accurate in a variety of domains [2]. Once the image is reconstructed, any appropriate image processing technique could be used to quantify features of the reconstructed image to estimate the same features in the full image. Alternatively, one could quantify the features in the subsampled images directly and scale those estimates to the full image appropriately (extrapolation). In this paper we further explore the bias and uncertainty introduced by the in-painting and extrapolation approaches to feature quantification of sub-sampled STEM images. Our goal is to provide STEM operators with the tools necessary to make informed decisions about how much dose to use when conducting their experiments as it relates to feature uncertainty. Several existing, high impact STEM experiments are used to illustrate our results [5].
Stanfill BA, SM Reehl, M Johnson, ND Browning, BL Mehdi, and LM Bramer. 2017. "Quantifying Feature Bias and Uncertainty Derived from Sub-sampled Low-Dose Scanning Transmission Electron Microscopy." In Quantifying Feature Bias and Uncertainty Derived from Sub-sampled Low-Dose Scanning Transmission Electron Microscopy.
| Pacific Northwest National Laboratory