Sparse Scanning Electron Microscopy and Deep Learning for Imaging and Segmentation of Neuron Structures

Tim Dahmen; Pavel Potocek; Patrick Trampert; Maurice Peemen; Remco Schoenmakers

In: Microscopy & Microanalysis 25. Microscopy & Microanalysis (M&M-2019), August 4-8, Portland, Oregon, USA, Cambridge University Press, 8/2019.


Sparse Scanning Electron Microscopy can be used in combination with Inpainting algorithms to reduce acquisition time and electron dose. In many situations, this approach leads to a higher image quality compared to images obtained by a conventional raster scan acquired at the same dose per pixel [1]. But what is the definition of image quality in this context? Human observers are unavoidably biased to prefer images that look good to humans. Most modern image quality metrics contain some “perception factor” as well that links the definition of image quality to the human visual systems. However, images are increasingly processed by machine learning systems such as Deep Neural Networks and evaluated automatically. In this context, the definition of image quality should be linked to the capability of a machine learning system to extract information from the image, not the human visual system.

sparse_scanning_electron_microscopy_and_deep_learning_for_imaging_and_segmentation_of_neuron_structures.pdf (pdf, 4 MB )

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence