Figure 2
From: Ice thickness monitoring for cryo-EM grids by interferometry imaging
ANN analysis of the interferometric images for the determination of ice quality. (a) Raw interferometric image of a copper grid coated with sample prior analysis by ANNICAS. Scale bar 50 µm. (b) The same image overlaid with the tiles (in grey) as recognized by the segmentation network after complete training (see main text and Methods). All tiles were correctly identified. (c) The same image as in (b) after full analysis and grid square classification with the fully trained classification network. The image was rotated during analysis. The grid squares are highlighted with colours according to the ice layer quality. Five classes for grid squares quality were defined: bad (class 0, in red), too thin (class 1, in white), optimal (class 2, in green), thicker (class 3, in bright green) and too thick (class 4, in orange), respectively (colour legend indicated on the right). Numbers indicate confidence of ANN. (d) Examples of grid square classification for a wide spectrum of different interferometric images. Ice thickness of grid squares ranged from ‘too thin’ (in white) to too thick (in orange), with also grid squares with ice thickness in between (in green or light green), or with grid squares presenting large defects (in red). Colour-code for ice quality shown on the top.
